No. 284 (September 2000)
Confused by metadata?
Following the publication in 1997 of the report of the EBU/SMPTE Task Force, the word "metadata" has become one of the most overused words in the vocabulary of broadcasting engineers. Despite Nicholas Negroponte’s wonderfully terse definition of metadata as "the bits about the bits", it is probably also one of the most misunderstood words. Simon Parnall (Chairman of the TV-Anytime Forum) recently observed that metadata is a Humpty Dumpty word: Humpty Dumpty is one of the characters in Lewis Carroll’s Alice through the looking glass who said, "When I use a word, it means just what I choose it to mean, neither more nor less".
It is important to differentiate between metadata used in the production process and metadata destined for use by consumers of broadcast material. It will become routine for cameras to record automatically the precise time, date and location of every shot, but such information is unlikely to be of great interest to viewers of a completed programme. Consumers (or their intelligent agents) will need a different set of metadata, notably information to aid selection of programmes of particular interest to them. As there will be overlaps between "production" metadata and "broadcast" metadata, it would be a mistake to develop systems in isolation from each other. On the other hand, it is not necessary to devise a unified system that will simultaneously satisfy all potential requirements.
In practice, metadata is not a new concept. Even in the "old-fashioned" analogue world, most videotapes had accompanying metadata, such as details about the duration of the programme or even detailed scripts. However, this information was generally written on a label or attached by means of an elastic band. In the digital world, such metadata will be stored along with the audio and video signals on tape or on servers.
Efficient exploitation of broadcasters’ archives will increasingly depend on accurate metadata. One of my repeated challenges to archivists has been whether their indexing systems can locate an aerial shot of Sydney Harbour Bridge at sunset. Rather than relying on sophisticated computer systems, it is surprising to discover that the success of such searches often depends on the memory of an individual archivist. In some cases, it may be cheaper to hire a helicopter and shoot some new material over Sydney! In the new world of comprehensive metadata, such searches "should" be easy.
This issue of EBU Technical Review contains three articles about metadata written from the perspectives of three individual experts. These articles outline progress in the varied ongoing attempts to develop specifications for metadata.
Metadata systems are being investigated by numerous groups including MPEG-7, the SMPTE and the TV-Anytime Forum, as well as by EBU Project Groups P/META and P/FTA (Future Television Archives). In addition, we need to be aware of parallel developments, such as the Dublin Core, which could have great value for broadcasters.
It is improbable that all of the interested parties across the various industries will be able to agree on harmonized systems for metadata. If we take the notionally simple task of defining programme types, it is chastening to realize that the EBU and its Members have already "completed" this task on several occasions. Programme types are listed in the specifications for RDS, DAB, MAC, DVB, PDC and analogue EPGs. Despite this work, such lists of programme types do not seem to have been widely implemented. In the case of TV systems, the lists were restricted to a maximum of 255 programme types. At first sight, this may seem sufficient but it means that a maximum of 16 categories are available to describe TV coverage of sports events. The resulting compromises are barely satisfactory – as can be seen from the following list:
Rather than naming individual sports, another approach is to adopt a hierarchical "tree" structure, such as:
| Sport | ® Ball games |
® Instrumental
ball games
|
| ® Non-instrumental ball games |
In principle, this concept is attractive – but the practical meaning of the above example is not immediately obvious to most people (including engineers). The category of instrumental ball games includes tennis and hockey as both games are played with instruments: respectively, a tennis racquet and a hockey stick! On the other hand, non-instrumental ball games include pétanque and football – even though few people play football without wearing football boots!
To avoid such problems, future systems for metadata must allow for more than 255 programme types. A more far-sighted approach is illustrated by the EBU ESCORT tables, which were agreed by the EBU Statistics Group in 1995 and have recently been "rediscovered" by the metadata community. Nevertheless, even if we could list a very large number of programme types, we would still be faced with intractable problems, such as which programme type includes "Big Brother"? It is essential that modifications to the lists of programme types can be extended and updated by downloading new data to receivers.
For many years, librarians have been struggling with the problem of how to define the subjects of books – with huge battles between supporters of, for example, the Dewey Universal Decimal and the Library of Congress classification schemes. Although we can take some comfort from the fact that these experts have not found an ideal solution, it illustrates that a very large tree structure does not necessarily solve all of the problems.
The danger with an unbounded problem as complex as that of metadata is that the experts will strive to find the perfect solution – and miss the window of opportunity for early implementation of less-than-perfect solutions. Surely, our plea to the metadata experts must be "sooner is much better than later".
Philip Laven
Director
EBU Technical Department
| European Broadcasting Union Case postale 45 Ancienne Route 17A CH-1218 Grand-Saconnex Geneva Switzerland techreview@ebu.ch |
 |