COLLATE

1
COLLATE Collaboratory for Annotation, Indexing
and Retrieval of Digitized Historical Archive
MaterialDELOS International Cooperation
Workshop, May 30, 2003

• Ingo Frommholz
• Fraunhofer IPSI, Darmstadt
• frommholz_at_ipsi.fraunhofer.de
• http//ipsi.fraunhofer.de/

2
Digital Libraries in Cultural Heritage

• Valuable historic document collections exist, but
  are scattered in national archives
• Sources mostly not available online
• Difficult-to-use database referencing systems
• Lack of content-based indexing access
• Valuable expert domain knowledge exists, but
  mostly inaccessible to externals
• Tacit knowledge, insufficiently documented
• Professional communities lack technology support
  for collaborative knowledge working

3
The COLLATE Project (IST-1999-20882)

• Constructing a Collaborative Information Space
• Preserve historic documents in a distributed
  multimedia repository
• European historic film documentation (20ies and
  30ies)
• Historic film censorship (legal docs,
  applications decisions, correspondence, etc.),
  Press material (articles), Photos (stills,
  portraits) film posters, Digital film/video
  fragments
• XML metadata (cataloguing content indexing)
• Ensure accessibility
• Work environment for content indexing
  annotation
• Content- and context-based retrieval
• Evaluate acceptability
• Preservation case studies by film experts
• Empirical studies of real-life user behavior

4
Partners

• Content providers / pilot users
• Deutsches Filminstitut DIF, Frankfurt, Germany
• Filmarchiv Austria, Vienna, Austria
• Národní Filmový Archiv, Prague, Czechia
• Technology developers
• Fraunhofer IPSI, Darmstadt, Germany
• University of Bari, LACAM Lab, Bari, Italy
• Sword ICT S.r.l., Bari, Italy
• Evaluation partner
• Risø National Laboratory, Systems Analysis Dept,
  Denmark

5
Why a Cultural Collaboratory?

• Support existing work processes in cultural
  sciences
• Interpretative content analysis of documents
• Reconstruct unity of cultural phenomena,
  interlinking scattered knowledge sources
• Offer new knowledge working environment
• Organize collaborative work
• Bring together divergent user communities roles
• Create enhanced cultural information services
• Raise awareness visibility of cultural archives

6
Censorship / Registration Cards
7
Newspaper Articles
8
Conceptual Integration COLLATE-Ontology
Collate Entity
Location
Generic Level ABC-Model
Temporality
Abstraction
Actuality
Cultural Heritage Do-main Level CIDOC CRM, FRBR
Manifes-tation
Work
Situation
Action
Event
Agent
Film Archive Subdomain Level LC TGM II FIAF
Classification
Form, Genre Physical Cha-racteristics
Moving Image
Film Situation
Film Event
Film Activity
Film Agent
COLLATE Appli-cation Level Collate Keywords
Film Censorship Agent
Film Cen-sorship Activity
Film Censor-ship Event
Censorship Document
Film- and Censorship Topic
9
Model of the Concept Film Life Cycle
hasParticipant
hasParticipant
cencorshipdecisionx
Directingx
hasAction
hasAction
shortedversionx
censor-shipx
filmcreationx
originalversionx
precedes
precedes
follows
has Result
involves
has Result
Filmcopy Bx
Filmcopy Ax
Work x
realizesWork
realizesWork
10
System Architecture (OAIS)
11
Collaboration in COLLATE
12
Discourse Structures

• Discourses represent extended communication
  between two or more participants in a shared
  context. (Rich Sidner, 1998)
• Establishing a discourse context
• Modeling discourse as interrelated nested
  annotations
• Annotation thread reflects scientific discourse
• Typed links (DSR) between
• Document and annotation
• Annotation of annotations

13
Communication Acts Discourse Structure Relations
14
Semantic Web Integration COLLATE RDF(S)
15
Document Retrieval in COLLATE

• For a query q, a ranking of documents is
  returned. Therefore, a retrieval weight r is
  calculated for each document.
• Documents are ranked according to descending
  retrieval weights
• The retrieval is based on the documents metadata
  (given by film scientists or extracted from the
  digitized documents) and on the annotation
  thread.

16
Context-based Retrieval in COLLATE

• In COLLATE, we deal with the discourse context.
• A document is seen in the light of its
  interpretations
• We also consider at which point of the discourse
  a statement is made and what relation exists
  between the statement and the entity this
  statement refers to.
• Example Consider the query for all censorship
  decisions made for political reasons.

17
Query censorship decisions for political
reasonsMetadata Only
... ltcontrolled_keywordgt obscene
actions lt/controlled_keywordgt ...
I think the reasons mentioned here are not the
real reasons. I see a political background as
the main reason.
Keyword
0.01
Inter- pretation
Counterargument
Document Interpretation
I disagree. There were a lot of similar decisions
with the same argumentation. Of course, there
might be a political background, but I think this
is not the main reason in this case.
Cataloguing
... ltassessors_chairmangt Oberregierungsrat Dr
Becker Beisitzer Justizrat Dr.
Rosenthal... lt/assessors_chairmangt ...
... ltfilmtitlegt Kuhle Wampe lt/filmtitlegt ...
18
Query censorship decisions for political
reasonsMetadata Interpretation
... ltcontrolled_keywordgt obscene
actions lt/controlled_keywordgt ...
I think the reasons mentioned here are not the
real reasons. I see a political background as
the main reason.
Keyword
0.32
Inter- pretation
Counterargument
Document Interpretation
I disagree. There were a lot of similar decisions
with the same argumentation. Of course, there
might be a political background, but I think this
is not the main reason in this case.
Cataloguing
... ltassessors_chairmangt Oberregierungsrat Dr
Becker Beisitzer Justizrat Dr.
Rosenthal... lt/assessors_chairmangt ...
... ltfilmtitlegt Kuhle Wampe lt/filmtitlegt ...
19
Query censorship decisions for political
reasonsAnalysis of Discourse Structure
Relations
... ltcontrolled_keywordgt obscene
actions lt/controlled_keywordgt ...
I think the reasons mentioned here are not the
real reasons. I see a political background as
the main reason.
Keyword
0.19
Inter- pretation
Counterargument
Document Interpretation
I disagree. There were a lot of similar decisions
with the same argumentation. Of course, there
might be a political background, but I think this
is not the main reason in this case.
Cataloguing
... ltassessors_chairmangt Oberregierungsrat Dr
Becker Beisitzer Justizrat Dr.
Rosenthal... lt/assessors_chairmangt ...
... ltfilmtitlegt Kuhle Wampe lt/filmtitlegt ...
20
COLLATE User Interface
21
Current State

• A first prototype was delivered to the archives
  and is used by them
• A second prototype will be delivered soon,
  introducing discourse structure relations and
  advanced collaboration features to the users
• A third prototype will contain context-based
  retrieval

22
Outlook

• Evaluate collaborative approach and context-based
  retrieval
• Apply COLLATE technology in other domains?

23
More information? http//www.collate.de