Proxy data objects

1
Proxy data objects
provide optional linking to external objects and
reusable data interfaces, from which Darwin- and
LinneanCore-like protocol interfaces can be
constructed
(Talk held Wednesday, 2004-10-13, at the TDWG
2004 meeting in Christchurch, New Zealand)
2
(Submitted Abstract)

• Ideally, biodiversity data are expressed using
  well-defined object types (with generally
  accepted and intuitive property and object
  composition concepts) and all objects required in
  relations are available in digital form,
  identifiable through resolvable globally unique
  identifiers. Reality is different. a) the complex
  models like ABCD, SDD, TCS etc. are under debate
  and not necessarily fully stable. A consequence
  is that simplified "Cores" like DarwinCore or
  LinneanCore are proposed. b) Most required data
  are not digitized at all. Wherever data should
  naturally refer to other biodiversity domains
  alternative models are required to either linking
  something external or provide a sufficient
  internal object definition. Such a definition
  also involves a simplified set of core elements.
• I therefore propose to combine the two problems
  and define relatively simple data interfaces,
  that can serve both for protocol/query purposes
  and for the definition of local proxy objects (
  either link to external entities, or provide a
  local definition). Data interfaces shield the
  complexity of a fuller model (i.e. the full model
  can be treated as a black box). They should be
  rough enough to fit to several models, but also
  detailed enough to allow the definition of proxy
  data (i. e. make a substantial semantic
  definition). Data interface are often implicitly
  used in current practice. 'Specify' uses a
  simplified literature and nomenclature interface,
  DarwinCore contains name, identification and
  geographical location interfaces, Taxon Concept
  Schema contains interfaces for literature and
  specimen, and Linnean Core has a literature
  interface. Agreeing on a common set of such
  interface concepts would allow to build Darwin
  and LinneanCore as well as much of SDD, TCS, etc.
  from a the same building blocks, drastically
  reducing the investment needed for a full global
  biodiversity information system.

3
Outline

• Linking and proxy objects
• Data interface definitions
• DarwinCore and LinneanCore from Interfaces

4
Linking

• Organism-interaction data are primarily a
  4-tuple of
• links to external objects
• Organism name 1 (e. g., a fungus)
• Organism name 2 (e. g., a plant)
• Geographical location
• Publication / data source reference
• plus non-linking data
• Interaction type (controlled vocabulary)
• Reference detail (page, document-fragment, ...)

5
Linking

• Similarly, SDD descriptions need links to
  express
• Taxon / class name
• Specimen unit
• Geographical scope
• Publication
• Contributing IPR agents
• Plus terminology definitions multiple
  description projects sharing a common terminology

6
Two kind of links?

• Convenience links (see also-links)
• GenBank entries for a specimen/taxon name
• Link to illustrating image/video
• If link breaks, the information content of the
  main document is usually left intact.
• Defining links
• Link to taxon / specimen in a description
• Link to cited publication
• If link breaks, the information content of the
  main document is lost or severely damaged.

? Recovery mechanism is highly desirable ?
Simplest preserve some human-readable semantics
7
Requirements

• A link should be
• Stable
• URLs are not stable
• PURLs unmanageable? (usually just xURLs
  extended life URLs)
• Resolvable
• Many GUID / URN schemes are not
• LSIDs and are both stable and resolvable
• Links used in primary scientific data must
  further
• Offer recovery mechanism if a (principally
  stable) link vanishes nevertheless!

8
Not required, but desirable

• Object Identity
• DOI defines identity, but is expensive
• At least, it should be discoverable, whether a
  link defines object identity or not ( do
  multiple URL / LSID exist for the same object?)

9
The Problem

• For quite some time, the default will be that
  there is no external object
• Either not yet
• Digitization of specimens
• Older literature
• Or unreliable
• Recent literature
• Moreover, permanently objects will be
  temporarily not yet available
• Science creates new
• Specimens still in private collection
• Taxon names / concepts to be published
• etc.

10
So, now we have
Linkclient
Linkedobject
Local recove- ry semantics
Or
(unavailable)
Linkingclient
Linkedobject
Local replacement
11
Often also caching desirable
Locally cached data
Linkclient
Linkedobject
Local recove- ry semantics
Caching linked data temporarily unavailable
Recovery linked data permanently unavailable
12
Simplify?
Linkclient
Linkedobject
Locally cached interface data (incl. recovery)
Proxy object
13
A data proxy

• May link to external data providers, especially
  for knowledge domains outside of the scope of the
  current dataset
• Supports several object linking mechanisms
  involving globally unique identifiers and
  resolving mechanisms (e. g., DOI, LSID, URL)
• Can replace links in cases where objects are
  (perhaps not yet) available from an external data
  source
• For existing links A minimalized data interface
  is cached on the assumption that access is
  asynchronous, slow, or may be temporarily
  unavailable
• For local proxies The same data interface allows
  relatively simple, local object definition to
  decouple processes
• Provides cached data and semantics to human
  readers, allowing recovery even if a link has
  become permanently broken

14
Outline

• Linking and proxy objects
• Data interface definitions
• DarwinCore and LinneanCore from Interfaces

15
Interface good choice of term?

• Not user-interface!
• Interface is used similar to its use in
  object-oriented programming
• However, no methods in data interfaces
• Persistence interface properties/fields/struc
  tures like collections
• Data nterfaces provide additional abstraction
  layer on top of public object model
• Data interfaces allow programming against the
  interface instead of against the full object
  model
• Is there a better term?

16
Complexity

• Complex schemata arecertainly necessary, but no
  solution for projects where a knowledge domain
  is of secondary relevance!

17
Goals

• Abstract encapsulate complexity
• Reduced size
• cover 80 of interaction needs
• react more flexible when changes in the main
  object model occur
• Formalize (and standardize) reality
• Most software needs peripheral objects from other
  knowledge domains and treats them lightly
• TCS, SDD, ABCD/DarwinCore, LinneanCore
• Example Specify contains a name and publication
  reduced data set with editors to them
  essentially proxy objects with a set of interface
  fields

18
Abstraction layers
Complex object model/schema
19
Data interface requirements

• Should be stable, so applications programming
  against the interface do not break
• Define mapping from complex standardto interface
• Where interface is used as proxy (entering data),
  the reverse should also be defined

20
Outline

• Linking and proxy objects
• Data interface definitions
• DarwinCore and LinneanCore from Interfaces

21
Modular interface/protocol schemata
Darwin-Core
Taxon name interface(full/atomizedtaxon name)
Taxon name interface(full/atomizedtaxon name)
Linn-ean-Core
Specimen curatorial interface(Collection/subcoll
.,access. no., )
Specimen curatorial interface(Collection/subcoll
.,access. no., )
Publicationinterface(author, year, title, ...)
Geographical location interface(place
description, geogr. coord., gazetteer link, )
22
Full schema partly using interfaces
Fulltaxonconceptschema
Taxon name interface(full/atomizedtaxon name)
Taxon name complete
Linn-ean-Core
Specimen curatorial interface(Collection/subcoll
.,access. no., )
Specimen curatorial interface(Collection/subcoll
.,access. no., )
Publicationinterface(author, year, title, ...)
Publicationinterface(author, year, title, ...)
23
OO schema or flattened?
Darwin-Core
Geographical location interface(place
description, geogr. coord., gazetteer link, )
24
Proxies proposed in UBIF/SDD