9 April 2000

Daniel Rivers-Moore
Director of New Technologies, RivCom
Consultant to the NewsML development process

Copyright © 2000 International Press Telecommunications Council
All Rights Reserved

This Working Draft is the first formal deliverable of the NewsML consultancy process, set in motion at the IPTC Spring Meeting in Nice, France, on 24 March 2000. It represents work in progress towards a full functional specification of NewsML. In its current form, its contents are not binding on IPTC or any of IPTC’s members.

The purpose of this document is to establish a baseline functional specification from which further work on NewsML can proceed. The document is deliberately neutral on certain issues that are under active debate and which do not need to be resolved until a later stage in the NewsML development process. However, it makes explicit statements of choice on issues where one or both of the following applies:

• There is sufficient consensus on the issue to warrant an explicit statement at this stage.
• An explicit statement on the issue is required in order to establish a firm foundation on which further technical work can be based.

A second Working Draft of the NewsML Functional Specification will be made available to the IPTC membership no later than the Working Party meetings due to be held in London on 4 and 5 May 2000. Working Draft 2 of the NewsML Functional Specification will be sufficiently complete and explicit (including on matters left open-ended in the current Working Draft) for it to be a candidate final functional specification for NewsML Version 1.0

The development process from Working Draft 1 to Working Draft 2 of the NewsML Functional Specification is as follows:

• Members of the CSC will comment on Working Draft 1.
• Working Draft 1 will be made available to the IPTC membership for comment and on-line discussion.
• The consultant will begin an assessment of which existing bodies of work (including but not limited to formal specifications or standards), within or outside the IPTC, may be directly relevant to the requirements of NewsML and how, if at all, these may be used in support of NewsML development (see below, the section entitled "Other bodies of work" for more on this topic).
• Based on the comments of the CSC and IPTC membership, and the consultant’s findings in relation to external bodies of work, the consultant will prepare a proposed Working Draft 2 of the NewsML Functional Specification, in which some or all of the issues left open-ended in this document will have been made explicit
• The consultant will prepare an Interim Report which he will present, together with the proposed Working Draft 2 of the NewsML Functional Specification, at a meeting of the CSC to be held in Swindon, UK, on 27 and 28 April 2000. The Interim Report will contain recommendations as to which external bodies of work are indeed relevant and useful for NewsML, how they can be made use of, and how in particular they have been used to render Working Draft 2 of the NewsML Functional Specification more explicit and complete than was Working Draft 1. The report will also map out the remaining stages of the development of NewsML itself, based on the existing NewsML Requirements and NewsML Encoding Decisions documents, the NewsML Functional Specification itself, and the ongoing work of the relevant IPTC Working Parties.
• After the incorporation of any amendments arising from the discussion at the CSC meeting of 27 and 28 April, the Interim Report and Working Draft 2 of the NewsML Functional Specification will be made available to the IPTC membership no later than the Working Party meetings due to be held in London on 4 and 5 May 2000.

The NewsML Functional Specification, together with the NewsML Requirements and NewsML Encoding Decisions documents, will provide a clear formal statement of the consensus reached by the IPTC membership as to the basis on which the NewsML Version 1.0 will be built:

• NewsML Requirements specifies what NewsML should do, or enable to be done. It does not limit itself to NewsML Version 1.0, but covers what NewsML should make possible in principle, even if certain aspects of this will not in practice be achieved until versions of NewsML beyond 1.0.
• The NewsML Functional Specification will specify how NewsML will meet those requirements. It will indicate how NewsML documents will be structured, how relationships between them will be established, how they can make reference to, and be referred to by, other XML-encoded resources and, to the degree to which this is achievable in practice, non-XML-encoded resources. It will spell out which subset of the identified requirements are to be met by NewsML Version 1.0, and, in general terms, how those not met in Version 1.0 may be addressed in subsequent extensions to, or versions of, NewsML.
• NewsML Encoding Decisions sets out a number of design principles that will guide the specific choices of syntactic mechanism to be used in meeting the NewsML requirements.

It is in the nature of design principles that they are intended to provide clear guidelines to the designers of a specification, rather than immutable rules. Thus, it is possible that the NewsML Functional Specification will in some cases depart from strict adherence to the principles laid out in the NewsML Encoding Decisions. However, any cases where this occurs should be explicitly recognised as exceptions to the principles in question, and the NewsML Functional Specification should include explicit justification for any such exceptions. Possible justifications might include considerations of comprehensibility, implementation efficiency, performance optimisation, ease of use of relevant tools (such as XML editing tools), or compliance with an external standard or specification. Such exceptions should (by definition) be rare.

The NewsML Functional Specification will include clear definitions of the specialised terms it uses, whether these be the formal names of constructs such as XML elements or attributes, or other terms used in explanations of those constructs and their intended use.

Working Draft 1 of NewsML Functional Specification covers the overall architecture of NewsML. It addresses issues of how NewsML documents will be structured, how relationships between them will be established, how they can make reference to, and be referred to by, other resources, and how, in general terms, extensions to and further developments of NewsML can be handled.

Working Draft 1 of NewsML Functional Specification does not define specific XML elements and attributes, or make choices about the names these may have. Nor does it specify which external standards or specifications will be used, either by inclusion or by normative reference. In some instances, it gives indications of specific options in these areas; in others, it leaves such questions open. This is a deliberate choice, as the purpose of Working Draft 1 is to register the consensus so far achieved, and to make decisions on fundamental issues which must be resolved before further progress can be made. It is not the purpose of Working Draft 1 to resolve all issues that are still the subject of active debate, nor to prejudge the decisions of the Working Parties responsible for other aspects of NewsML, or the conclusions the consultant will reach when carrying out his survey of external specifications or bodies of work.

By contrast, Working Draft 2 of the NewsML Functional Specification will make explicit the proposed choices of XML encodings, including naming conventions and normative or other references to other specifications. It will however leave sufficient freedom of choice within the structures it lays down to allow the Metadata and News Management Working Parties to continue their work unconstrained with respect to the business purpose of their activity. In other words, the NewsML Functional Specification will state how news management information and other metadata will be structured, but it will not place bounds on which specific pieces of news management information or other metadata are to be supported by NewsML. These choices are the responsibility of the relevant Working Parties, not of the authors of the NewsML Functional Specification.

Other bodies of work that may be deemed relevant to NewsML include:

The ways in which such bodies of work might be used as a support for NewsML include:

• normative references to external standards within NewsML
• the definition within NewsML of means whereby NewsML documents may call upon the capabilities of external standards by inclusion or by reference
• the reuse by NewsML of some of the concepts lying behind other standards or specifications
• the application within NewsML of guidelines for good practice that are explicit or implicit within other bodies of work
• the use within the formal specification of NewsML itself of constructs defined by other formal specifications (Examples of this would be the inclusion, as part of the NewsML specification, of an object model defined using UML, or an XML document structure defined using XML Schema, or a mapping between two or more XML document structures defined using XSLT.).

The NewsML Requirements document sets out the capabilities that NewsML is required to deliver. This NewsML Functional Specification sets out the technical means that will be employed to meet those requirements. The requirements can be briefly summarised as follows (numbers in brackets preceded by the letter R are references to the relevant clauses in the NewsML Requirements document):

NewsML is to be an compact (R900), extensible and flexible (R700) structural framework for news, based on XML and other appropriate standards and specifications (R1000). It must support the representation of electronic news items, collections of such items, the relationships between them, and their associated metadata (R100). It must allow for the provision of multiple representations of the same information (R500), and handle arbitrary mixtures of media types, formats, languages and encodings (R300, R400). It must support all stages of the news lifecycle (R600) and allow the evolution of news items over time (R200). Though media-independent, NewsML will provide specific mechanisms for handling text (1100). It will allow for the authentication and signature of both metadata and news content (R800).

In the course of this overview, we shall introduce some terms that will be used in this document to describe the structure and design of NewsML. These terms are purely descriptive and should not be presumed to resemble any future formal names of NewsML constructs such as XML elements and attributes. To prevent any misunderstanding on this point, we have deliberately used composite terms such as ‘information carrier’ and ‘information content’, without concatenation, so they could not be possibly be used as XML names in their current form. These terms and their definitions are reproduced in the Glossary section at the end of this document.

NewsML is primarily intended as a format for the interchange of news, rather than its creation, editing, storage or in-house management. However, NewsML will be inherently extensible, and so can provide the basis for the development of further, specialised formats for these and other purposes. Such formats may be proprietary or may result from subsequent standardisation efforts.

The primary encoding used by NewsML will be XML. This simple statement carries with it several important consequences.

Firstly, it implies that the objects of NewsML are logical rather than physical. An XML document is a single logical object, though it may be built up of the contents of multiple physical files through the mechanism of XML entity references, for example.

With this in mind, we shall use the term data object to represent a logical unit of data, whether or not it is stored as a single file or transmitted as a single data stream.

Secondly, the use of XML implies that all the data-management capabilities that are provided by XML and its associated specifications are available to NewsML, and are inherited by it. In particular, the ability of XML to handle data in non-XML formats, through the NOTATION mechanism, is available to NewsML also. This powerful feature of XML will provide one possible way of addressing some of the media- and encoding-independence issues that are part of the NewsML requirements.

In order to be handled appropriately by an application or business process, a data object often needs additional data to be associated with it. We shall use the term metadata to designate data that is associated with a data object with the intent of enabling a system to handle that data object appropriately. The system may be a computer application, a business process handled by human beings, or some combination of the two. Thus, metadata may need to be machine-interpretable, human-readable, or both. A metadata object is a data object consisting of metadata.

NewsML makes no assumption about the media type, format or encoding of news. It provides a structure within which pieces of news media relate to each other. NewsML can equally contain text, video, audio, graphics, photos, or other media and combinations of media yet to be invented.

We shall use the term information carrier to designate a data object whose data content is intended to convey information (such as news content) to human recipients. Thus, the video component of a TV news report, the braille-encoded content of a newspaper article destined for blind readers, the electronic encoding of a table of stock-market closing prices, and the textual content of an article destined for a piece of newsprint, would all be described as information carriers.

We shall use the term information content to designate the information carried by an information carrier and intended for human consumption. (Note that we use the term ‘information content’ rather than ‘news content’ in order to make the NewsML specification as general-purpose as possible. A NewsML information carrier could for example be used to carry historical information intended for use in the biography of a historical character, or a documentary about a past event, as easily as it carries up-to-the-minute information about current events.)

It is important that NewsML cater to the needs of traditional formats as well as new or emerging formats. Many information carriers contain structured or unstructured text. Care must therefore be taken to ensure that allowing NewsML to handle non-text-based information does not result in a reduction in the ease and efficiency of text-based news handling.

It is frequently necessary to associate a data object with one or more others, which may be physically close to or physically remote from it. Furthermore, the data that encodes the association between data objects may be internal to one or both objects, physically close to one or both, or physically remote from both. The same is true of the metadata associated with a data object: some or all of the metadata may be physically remote from the object to which it relates.

We shall use the term association information to designate the information that conveys the existence and nature of an association between any number of data objects and/or metadata objects, wherever that information is physically held.

We shall use the term association data object to designate any data object that encodes association information.

It is possible for association information to exist without there being any corresponding association data. For example, the fact that a given XML element is the sibling of another will often imply the existence of a meaningful relationship between those two elements. This relationship counts as association information despite the fact that there is no association data that represents the existence of this relationship, beyond the structural relationship that pertains between the elements themselves. We shall use the term implicit association information to designate association information that is implicit in the structural relationships between the data objects themselves rather than being represented by association data objects.

Like any other data objects, association data objects can themselves have associated metadata. We shall use the term association metadata to designate metadata associated with association data objects. Examples of association metadata would be the date at which an association data object was created in order to assert a relationship between two data objects, the name of an editor who created the association data object, the reason why it was created, a date before which it is not deemed to be valid, etc.

We shall use the term information object to designate any data object plus its associated metadata. Another way of putting this is to say that an information object is a data object that is able to be used appropriately by a system. This follows from the fact that metadata is defined as data associated with a data object with the intent of enabling a system to handle it appropriately.

It is a characteristic feature of news stories that they often bring together multiple information objects, for example, a text story, a photograph and its caption, and a vector graphic. Further, it is often necessary to bring together multiple complete stories and handle them as a coherent collection, for example in a digest of the week’s major stories, or as a response to a query seeking out stories relating to a particular event or theme.

We shall use the term information collection to designate an information object comprising any number of information objects that are intended to be used together for any purpose. The components of an information collection are zero or more information objects (data objects plus associated metadata), at least one additional item of metadata associated with the information collection itself, and any relevant association information defining the relationships between the information objects that make up the collection. Just as an information collection is itself an information object, so any or all of its component information objects may themselves be information collections.

Special cases of information collection are ‘the collection of all versions of an article as it proceeds through the editorial workflow from authoring, to editing, to publication’, and ‘the collection of all revisions of a story as new facts are discovered and the story is refined and augmented’. Similarly, special cases of association information are the fact that one information object is an earlier version of another, and a correction or replacement to a third. NewsML will provide explicit support for these kinds of situation through the provision of specific categories of metadata and association information appropriate to the management of the entire news lifecycle.

An information collection is not necessarily intended to be used in its entirety at any given moment of news delivery. It might for example include multiple alternative versions of some of its component information objects. The fact that one information object is an alternative to another within the context of a particular information collection is represented by appropriate association information specifying the relationships between the relevant objects.

Criteria for the choice between alternatives might relate to any aspect of the metadata associated with the objects. The choice might depend, for example on the format (e.g. video, audio or photograph), the encoding (e.g. JPEG, EPS or SVG graphic), the available space (choice between short or long headline, large or small photograph, etc), the language in which the textual component of the information is written, the language spoken by the speaker in an audio clip, or any other consideration dependent on the constraints or requirements of the delivery environment, process or intended audience.

As has been explained above, information objects and information collections are logical rather than physical objects. However, it is sometimes necessary (for example for data transmission or data storage purposes) to render the content of these logical objects in a single file or data stream. We shall use the term information package to designate a physical file or data stream that is used to store or transmit one or more information objects or information collections.

The various protocols or security features of the data storage or transmission environment might require certain information to be included in the information package in addition to the actual information objects it contains. We shall use the term information envelope to designate this additional information. An information package therefore comprises an information envelope together with the information object(s) or information collection(s) that are being stored or transmitted.

We shall use the term explicit inclusion to designate the physical inclusion of the entire data content of an information object within an information package. We shall use the term inclusion by reference when an information package does not physically include the data content of an information object but instead includes a pointer or other form of reference data, sufficient to enable a NewsML-aware system to retrieve the actual information object at some later point in time, if it is required.

NewsML provides the means for both explicit inclusion and inclusion by reference to be used at will. The criteria by which a given application decides whether to use one inclusion mechanism or the other, or some combination of the two, are implementation-dependent. Senders and receivers of NewsML packages are free to negotiate their own rules in this regard, in the light of the specific constraints or requirements of their physical, technical or business environments.

NewsML will provide a generic information carrier object, whose purpose is to represent any item of news-related information, destined for human consumption, through some data encoding. The eventual presentation of the information to a human recipient could be through any medium – recorded sound, written or speech-synthesized text, printed or on-screen photograph, or other media yet to be invented. The rendering of the information to the recipient will be through some physical device that is able to interpret the encoding in which the information is carried.

The information carrier on its own is simply a raw data object without any meaning beyond that implicit in the semantics of its encoding, and even that is of little use unless it is accompanied by additional information to tell the system what that encoding is. The information carrier becomes an information object through the addition of appropriate metadata, which conveys all the additional information relevant to telling a system how the object is intended to be used.

NewsML will specify a number of different kinds of metadata, each relating to a different aspect of the ways information objects may need to be used within NewsML-aware systems.

We shall use the term metadata category to designate a broad class of metadata objects that it is useful to consider as a group because the objects that fall within that category are relevant to a particular aspect of the business purpose or practical use to which NewsML information objects may be put.

We shall use the term metadata token to designate a metadata object of a particular kind, falling within one of the supported metadata categories.

We shall use the term metadata property to designate that aspect of an information object that is represented by a particular metadata token; put another way, a metadata property is a class of information conveyed by metadata tokens of a particular type.

To give an example: within the metadata category known as ‘physical’, whose purpose includes the description of certain aspects of the way an information object will be rendered, will be a metadata token called, perhaps ‘width’, which represents a particular property of information objects, namely the amount of horizontal space they will occupy when rendered.

It should be noted that metadata properties may themselves be structured. It cannot be assumed that a single XML attribute, for example, will be sufficient to carry the semantics of a metadata property. In the example given, the metadata property known as ‘width’ is not atomic. As a minimum it must be expressed as a numeric value accompanied by a unit of measure (for example, ‘5 inches’ or, equivalently, ‘12.7 centimetres’), not just as an unadorned numeric value. In fact, it may be yet more complex, as it might be decided to allow width to have multiple possible values such as ‘recommended with’ and ‘minimum width’, within a single, complex metadata token representing a similarly complex metadata property comprising a set of related widths, each with a different role.

Note: At this point in the NewsML Functional Specification will appear a list of the relevant categories of metadata together with a brief description of the purpose of each. At the time of writing Working Draft 1, this list is not definitively agreed. At least two candidate lists have been proposed. One (included in the Reuters draft functional specification) consists of five categories: descriptive, editorial, identification, role and physical. Another (included in the AFP draft functional specification) consists of seven categories: identification, administrative, descriptive, physical, roles, rights and management. It will be for the relevant IPTC Working Parties to make appropriate decisions as to the definitive list to be adopted by NewsML Version 1.0, and to spell out the specific metadata objects that will be able to exist within each category.

There is no theoretical limit to the kinds of use to which information objects may be put, and consequently the metadata categories that may be required to provide information relevant to those kinds of use.

It is clear that certain kinds of information object require certain kinds of metadata and not others. An audio object does not have a ‘width’ property, for example, and so does not require a ‘width’ token to encode it.

We shall use the term information object type to designate a class of information objects to which certain metadata categories are relevant.

It is clear from the above list of metadata categories that the scope of NewsML metadata is extremely broad. It is therefore important that the NewsML specification define a special metadata category comprising the minimum subset of metadata properties that all NewsML information objects must have, and to specify among those which are ‘required’ and which are ‘optional’. We shall use the term minimum required metadata to designate this subset.

For those metadata properties that are deemed to be ‘required’, the corresponding metadata tokens may in fact remain optional, since NewsML may specify default values they are deemed to have if not physically present in a NewsML data object. A required metadata token will be one corresponding to a required metadata property for which no default value has been specified. The default values may be specified in the NewsML specification itself, or in an agreement between the sender and receiver of NewsML information objects.

The NewsML specification will include an explicit list of metadata properties within each metadata category. Certain metadata properties will be allowed any value at all, provided it is of an appropriate data type. Others may require the value to be within a certain range (that is, between specified minimum and maximum values). Others may require the values to be drawn form an explicit list of permitted values, known as a controlled vocabulary.

We shall use the term controlled vocabulary to designate a list of permitted terms that may be used for a particular purpose. The content of a controlled vocabulary is not fixed. However, changes to it can only be made according to a specified and well-managed process.

If a given metadata property needs a specified list of permitted values, this list will take the form of a controlled vocabulary.

There may be more than one controlled vocabulary that can be used for a given purpose within NewsML.

NewsML will provide a mechanism for specifying which vocabulary is being used in any particular instance. Thus, for example, in specifying the value of a particular metadata property, it will be possible

• to specify the value
• to identify which controlled vocabulary this value is drawn from
• where appropriate, to identify which revision of the controlled vocabulary is being used
• to identify the body responsible for maintaining the controlled vocabulary
• where appropriate, to provide a link or reference (perhaps in the form of a URI) to the controlled vocabulary itself, or to some documentation relevant to it.

NewsML will provide a formal mechanism for asserting that the meaning of a term in one controlled vocabulary and the meaning of a term in another controlled vocabulary are the same. These terms will then be considered to be equivalent whenever they are used within NewsML information objects. Likewise, where terms in controlled vocabularies designate objects in the real world (such as places or countries), there will be a mechanism for asserting the identity of their referents. Thus, for example, it will be possible to assert that the town designated by the term ‘Lyons’ in a British controlled vocabulary is identical with the town designated by the term ‘Lyon’ in a French controlled vocabulary.

Just as it is necessary for the lists of permitted values of metadata properties to evolve over time in a controlled manner, so it is necessary to allow for new metadata properties to be recognised over time and supported by NewsML systems, even if they were not explicitly included in the NewsML Version 1.0 specification.

The relationship between a data object and its associated metadata objects is asserted through the existence of appropriate association information. This association information may be implicit or explicit.

Information objects may be composites consisting of multiple component information objects, each including its own metadata, and associated by implicit or explicit association information. The component information objects may in turn be information collections in their own right, to arbitrary levels of nesting.

The ultimate information carriers that are contained in the lowest-level information objects in the collection may be of any media type and may be in any encoding. NewsML will include a specific metadata category whose intent will be to specify the encoding and other physical characteristics of each information carrier, so that it can be appropriately selected, decoded and rendered by a NewsML-aware system.

NewsML will provide a simple mechanism for creating an information collection or information package by specifying an entry point into a web of associated information objects, and providing a set of criteria whereby, starting from the information object designated as entry point, associated information objects are included, explicitly or by reference, within the collection or package.

The process whereby the collection or package will be built up is to locate the association information relevant to the initial information object, apply the specified criteria to determine which associations are to be followed and which not, then to repeat the process for each information object thus located, and so on, recursively, until the rules being followed identify no further relevant associations.

NewsML will specify how alternative revisions and representations of the same news story shall be managed and related to one another.

It is recognised in the NewsML Requirements document that, while NewsML is itself media-neutral, nonetheless there are certain ways in which text plays a special role. It is a stated requirement of NewsML that it handle these text-oriented features efficiently and effectively.

The text-oriented features of NewsML include

• metadata properties of information objects, whose value is a structured or unstructured piece of text intended to provide information to a user about the nature of the information object itself. Examples of such text-oriented metadata properties are

• the caption of a photograph or graphic
• the headline of a newspaper article
• the by-line, credit-line or copyright statement of an information object of any media type
• a brief description of an information object of any media type, intended to be displayed in a pick-list from which a user may select the information object in question. (Note that, though this is a text-oriented metadata property, there is in fact a graphical equivalent, namely a thumbnail intended to be used in a picklist of graphical, video or multimedia objects.)

• Any information object whose encoding is a text-based format. These will include

• textual news content marked up in NITF
• textual news content marked up in XML or SGML, in accordance with a specific DTD or Schema (including any of the HTML DTDs, or XHTML)
• textual news content marked up in well-formed XML without a DTD
• an information object whose rendering will be graphical or multimedia, but whose encoding is textual (such as PostScript, cleartext EPS or PDF, SVG, VRML etc.).

We shall use the term content metadata to designate metadata whose purpose is to provide information about the information content of an information carrier, no matter what encoding the information carrier uses or what media-type is used to render it for human consumption.

Content metadata may occur under a number of different metadata categories. The most obvious example of call content metadata is what the Reuters and AFP draft functional specifications both designate by the term ‘descriptive metadata’. This metadata category is explicitly intended to convey information about the information content conveyed by an information carrier. However, this is not the only example of content metadata. In the case of a video of a news report, the journalist’s sign-off statement is part of the information content of the video clip. In this case, the metadata information objects (within the ‘editorial’ or ‘administrative’ metadata category) which identifies the journalist and the date and place of the report count as content metadata because they convey the same information as part of the information content of the information carrier itself.

NewsML will provide mechanisms for:

• specifying that a particular piece of metadata repeats a part of the information content
• specifying that a particular piece of metadata provides additional information about a part of the information content (it might for example identify, through the use of some scheme which will itself be identified, a specific person, place, event or organisation that is referred to within the information content)
• identifying (where this is appropriate and where the encoding of the information carrier is such as to make this possible) which ‘section’ of the information carrier carries the information that is repeated or amplified by the metadata in question.

There are many ways in which sections of an information carrier can be designated as being relevant to a particular piece of metadata. These are all dependent of the nature of the information carrier in question, and in some cases are also dependent on its encoding. Some examples are:

• an XPointer identifying a particular element, attribute or text span within an information carrier that is encoded in XML (Note that this may have any number of semantic implications. If the information carrier in question is an SVG graphic, for example, the element thus identified may in fact represent a particular part of a graphic, such as a circle surrounding the face of a person identified by a reference to a person identification scheme)
• a byte range identifying a substring within the string of characters that constitutes an information carrier encoded in plain ASCII text
• the coordinates of the bounding points of a shape within a graphical object
• the start and end times of a segment of an audio or video object
• an SQL query that returns a subset of the content of a tabular array of data in an information carrier encoded in CSV format.

NewsML places no limits on the formats or syntaxes that may be used to encode such pointers and references into parts of the information carriers it handles. It will provide mechanisms for stating what encoding is being used. As with the encodings of the information carriers themselves, NewsML may designate some preferred encodings, and/or some encodings that are deprecated, for such pointers. The mechanism itself is generic. The only fundamental restriction is that such an external pointer must be meaningful in relation to the encoding of the information carrier into which it points. In other words, it must be capable of performing its function of identifying a part of the information carrier to which a piece of content metadata relates.

NewsML will provide mechanisms for stating the identity of the person, computer system, organization or authority that creates, transmits or publishes any information object, whether it be an information carrier, a metadata object or an association information object.

It will therefore be possible for recipients of NewsML information packages to form judgements as to the confidence they place in the information they receive, based in part on the identity of the people and/or organisations from which it comes.

Furthermore, NewsML will provide mechanisms whereby a person or organisation may specify the degree of confidence they themselves place in a particular information object. Thus, it would be possible, for example, for a news agency to transmit a story while expressing doubts as to its veracity, or for a photo editor to identify a face in a photograph and state that they believe, but are not certain, that this is the face of a specified person.

This capability of NewsML can be used to attach statements of confidence (or lack of it) to parts of an information package that is transmitted to a customer, for example. It may also be used purely internally, as part of the process of deciding which stories to run, or as part of a research phase of the editorial process, where different stories are being corroborated against one another in order to assess the degree of confidence one wishes to place in them.

Given that the scope of NewsML metadata is considerable, and may grow as NewsML evolves, it would be inappropriate to require every NewsML system to support every NewsML metadata category. It is clearly of no relevance for a radio broadcasting system, for example, to support metadata relevant to the rendering of photographs. However, it is a requirement of every conformant NewsML system that it support at least the minimum required metadata as defined by the NewsML specification.