On The Evolution of Terms

"appellavitque Adam nominibus suis cuncta
animantia et universa volatilia caeli et
omnes bestias terrae..."

Genesis, 2:20

Orri Erling - Program Manager, Virtuoso
Yrjänä Rankka - Developer, Virtuoso

"Adam called all things by their names"

• We are not the first to try
• The history of the "search for the perfect language" favours natural languages
• What gets used becomes fit for the task
• Language building from scratch has generally not been successful
• Forcing nature to fit preconceived, idealized taxonomies has generally failed

Usage for Linked Data

• New layer to the document web
• Use de-referenceable HTTP URIs
• Use #this to distinguish subject matter from document
• Reuse terms where can
• Human readable URI's are best
• From HTML to XML, most formats are somewhat human readable, same for the data web

Universal "Data Language"?

• RDF will do for grammar
• Useful vocabulary can only evolve in a community of practice
• Distributed evolution of vocabulary and diversity of names for things is a given
• Application determines the data structure and vocabulary

Emerging Hubs

• SIOC/FOAF For Web Data Spaces
• DBpedia/UMBEL/OpenCYC for names of real world entities
• BFO for epistemology of structures and processes

Can't Put Genie Back in the Bottle

• Once data is out, there it stays
• Single identifiers for entities are possible only in closed, application specific DWs
• Heterogeneity of names, overlap of descriptions, taxonomies etc is a given
• Meaning exists only in context, so make this explicit

Infrastructure Scenarios

• Application-specific warehouse or mapped RDBMS
• General warehouse with lots of graphs, a la search engine or Billion Triples Challenge
• Query driven harvesting a la OpenLink Sponger
• On-line discovery and federated SPARQL

Implications for Query

• Make it explicit
• Report what data sets and SameAs's and graphs went into producing an answer
• Allow the app to explicitly choose what graphs, SameAs's, taxonomies, etc. are considered
• Search and discoverability will drive vocabulary convergence

Sameness

• SameAs and equivalent are necessary and permanent features
• What can be considered identical depends on context
• Universal agreement will not happen, so let people choose whose SameAs they trust
• SameAs adds query cost and must be resolved at time of query
• SameAs cannot be forward chained at web scale because which of them are relevant is not fixed
• Malicious/SPAM SameAs is inevitable

Implications For Publishers

• SPARQL and data self description practices are urgently needed
• Describe what names are used and what other data meshes with yours
• Explicit license

Implications for Entity NS

• DNS is good because of distributed, resilient storage and admin
• Convergence cannot be forced but should be encouraged
• Make administration compartmentalized a la DNS for No SPAM, No censorship
• People can say things in their own spaces
• Offer classification, e.g. UMBEL
• In searching for terms, rank most reused the highest

Conclusion

• Since uniformity is impossible, make diversity of identifiers explicit
• Build alongside the document web, de-referenceably and with #this
• Encourage reuse but allow innovation
• Needs of communities will differ according to stage of development

OpenLink Software

Thank You!

http://www.openlinksw.com