Towards the Semantic Web

• Berners-Lee vision of the "semantic Web"
• Virtual communities with shared semantics
• Need to develop shared understanding of data
• Semantics of relevance to museums, archives, libraries:
  • Metadata, e.g. OAI, ABC Harmony
  • Detailed records, e.g. SPECTRUM XML DTD for interchange
  • Authority information - "a suitable case for treatment"

What is a "controlled vocabulary"?

• A resource which provides a limited set of descriptors, and possibly a definition, for each concept of interest
• Often defines relationships between concepts
• Some types of controlled vocabulary:
  • Word list
  • Thesaurus
  • Classification system
• They are typically used:
  • To improve consistency of recording in cataloguing systems
  • In search interfaces

Controlled vocabularies - some questions

• Which of you uses controlled vocabularies in your work?
• What format are they in (printed, computer-based, ...)?
• What do you do with them?
• Who created them?
• Are they maintained and updated? If so, how?
• Have you tried to share your vocabularies with others?
• Have you tried to borrow someone else's vocabulary?

Features of word lists

• Often used to ensure consistency of recording
• There are no definitions: meaning is "understood" (!)
• There is no structure, just a list of terms
• Typically not used to aid retrieval
• Example: object names for MODES

Example: using a termlist for data validation

• A MODES data file that we want to edit
• The actual form of the termlist file
• Points of interest:
  • The termlist is held in a plain text file. Sometimes termlists are held "inside" your database system
  • The termlist is in a custom format, unique to MODES. Other software couldn't make sense of it

Features of thesauri

• Defined by an International Standard (ISO 2788 for monolingual thesauri; ISO 5964 for multilingual thesauri)
• A controlled vocabulary "formally organized so that a priori relationships between concepts ... are made explicit"
• Defines Broad Terms and Narrow Terms, leading to hierarchies of terms; and Related Terms
• Terms can appear in more than one context
• Used for both cataloguing and retrieval
• Example: mda Archaeological Objects thesaurus

Features of classification systems

• Concepts may require a detailed definition
• Often use numerical codes, instead of words, to represent concepts
• Tend to be a single hierarchy
• Numerical codes are less friendly than words
• Numerical codes give the impression (!) that mappings such as translation will be easier
• Example: Social History and Industrial Classification (SHIC)

Example: using controlled terms for searching

• A MODES web application which supports keyword searches
• Points of interest:
  • Access points need vocabulary control as well
  • Can be field names, or indexes
  • The names of indexes are important, and are not standardized
  • The terms for searching are derived from the data, not the termlist; other applications might allow the thesaurus or classification to be browsed
• mapping of SPECTRUM to Dublin Core

Updating and sharing vocabularies

• In principle it's a good idea. You can:
  • Share the work of developing the vocabulary
  • Achieve consistency between different museums (and even with archives and libraries ...)
  • Pave the way for searching across several museums (etc.)
• However, there is one big problem:
  • No standard format for delivering vocabularies!
• Revision of the ISO Standard for Thesauri may include a standard interchange format

Controlled vocabularies and XML

• XML offers a possible delivery format for controlled vocabularies:
  • XML is system- and platform-independent
  • XML is easily converted to other formats ("other" XML, HTML and "text") using XSLT
• Even if we adopt XML, we still need to develop an agreed encoding scheme
• XML could be used for:
  • Delivery of complete vocabularies
  • Delivery of updates to an existing vocabulary
  • Delivery of a single concept from within a vocabulary

Examples of controlled vocabularies encoded in XML

The XML in all of these examples has been generated dynamically from a database:

• Social History: SHIC
• Iconography: ICONCLASS: class 11, class 11A and class 1_11H
• Authority file: ITIS taxonomic authority: bobolink, common nighthawk
• Thesaurus: mda Archaeological Object Names thesaurus

Delivering consistent XML

• These examples all use different encoding strategies
• Conversion to an agreed XML standard for vocabularies can be automated, using XSLT for XML-to-XML conversion:
  • SHIC
  • ICONCLASS
• No need for individual suppliers to change their XML

Producing user-friendly documentation

• Once we have consistent XML, XSLT can (again) be used to transform it for:
  • printing
  • on-screen (HTML) help
• Examples:
  • SHIC
  • ICONCLASS

Updating the vocabularies in your database

• Use XSLT to convert the XML to "text" (i.e. to any required format)
• Updates can be presented in the appropriate format for loading directly into the database
• Each site (or at least each software package) will have its own requirements: develop a separate XSLT style sheet for each
• Example: SHIC data as "tag file" for loading into MODES

XML vocabularies on the Web

• Think beyond in-house vocabularies: why not put them on the Web?
• Many vocabularies are database-driven, and on the Web, in any case (delivering HTML): AAT term entry for "myths"
• Benefits of Web-based vocabularies:
  • Permanently available
  • Online data validation becomes possible
  • Vocabulary resources can be used to support searching
  • The updating problem goes away!
• The same approach can be used in-house (via a Web server on an intranet)

How a Web-based controlled vocabulary might work

• Merging local database with authority data: MODES and ITIS
• Demonstrates:
  • Searching against the authority on a non-unique key (common name)
  • Checking local data against the authority; reporting inconsistencies
  • Merging of data from the authority into the local report
• The XSLT style sheet that achieves this integration

The need for a "Web Query Language"

• Accessing XML resources on the Web requires a 'query language' (= HTTP request) as well as a 'delivery format' (= XML)
• A suitable HTTP request to the ITIS server will deliver the required XML, e.g.
  http://res.agr.ca:8800/itisca/owa/next?p_format=xml&taxa=&v_tsn=179032
• ... but you need to know the specific syntax that ITIS uses
• We need a standard HTTP syntax to use when accessing vocabulary resources (= another community venture!)

The importance of unique keys for concepts

• The MODES-ITIS example gave variable results (zero, one or more "hits"), searching on common name
• ITIS primary keys (TSNs) are unique, but not easy to guess unless you know them (or have retrieved the record already!)
• It's important to know whether the data you record is a unique key in the online authority
• Dilemma: unique key is often not human-friendly
• If authority always available, can pick up "friendly" info on the fly (e.g.SHIC data in the MODES web application)

Linking strategies

• 'Extended browsing': linking to related resources
  • Non-unique identifiers will do the job
  • 0 and multiple hits aren't a problem
• 'Information engineering': removal of redundancy by linking to external resources
  • Relies on recording unique, unchanging, identifiers (TSNs don't change, even when the taxon ceases to be current)
  • Aiming at precisely 1 hit per link

Cross-walks between different authorities

• Establish equivalences between concepts in two or more controlled vocabularies
• No need to alter the vocabularies themselves: use "out of line linking", e.g. XLink, Topic Maps
• Allows query expansion into another domain
• Potential technique for query translation (e.g. MACS project)

Conclusions

• Need an agreed XML syntax for controlled vocabularies and HTTP conventions for querying an online authority
• No need for authority XML to be standardised at source
• ITIS already available as XML: others (ICONCLASS) considering it; AAT would be nice! Need to encourage suppliers in this direction
• Authority can contain valuable information in its own right
• Can treat object data as an "authority" and achieve the same benefits
• Online XML authorities can be the "glue" in a museum semantic web

Some relevant projects

• AMP (Accès Multilingue au Patrimoine) - projects to enable multilingual access to Heritage information
• ABC Harmony (complex metadata; uses CIDOC CRM): http://metadata.net/harmony/index.html
• NKOS (networked knowledge organisation systems): http://nkos.slis.kent.edu/
• OAI mailing list (access via XML-based metadata standards): http://oaisrv.nsdl.cornell.edu/mailman/listinfo/oai-general.
• MACS prototype (interoperability of library classification systems): http://infolab.kub.nl/prj/macs/macsd.html

Some XSLT style sheets

• normalization of SHIC classifications
• normalization of ICONCLASS classifications
• displaying normalized classifications as HTML
• rendering SHIC classifications as a 'tag file'
• including SHIC authority information in a MODES record