Supporting the Generation of Argument Structure

1
Supporting the Generation of Argument
Structure within Video Sequences
2
Talk Outline

• The motivation and vision of the work
• What is needed
• Annotations
• Editing Process
• Editor Support
• Conclusions

3
Existing Documentaries

• Traditional video authoring the footage is
  selected and edited for the final cut
• there is only one final version, what is shown is
  the choice of the author / editor
• Material can be very rich and controversial (e.g.
  Voices of Iraq)

4
New Paradigm

• Proposed video authoring
• Annotate the video material semantics
• Edit it automatically, selecting what the user
  asks to see
• Use the Web as an interactive distribution mean
• More than a sequence of matching video
  fragments)
• Argumentation/rhetoric
• Narrative

5
Video material

• Interview with America video footage with
  interviews and background material about the
  opinion of American people after 9-11
• www.interviewwithamerica.com
• Annotations 1 hour annotated, 15 interviews, 60
  interview segments, 120 statements

6
What do you think of the war in Afghanistan?
7
Example Explained
Two billions dollar bombs on tents
contradict
I cannot think of a more effective solution
weaken
I am not a fan of military actions
support
War has never solved anything
8
The annotations

• Rhetorical
• Argumentation model Toulmin model
• Rhetorical Statement (mostly verbal, but visual
  also possible)
• Descriptive
• Question asked
• Interviewee (social)
• Filmic (e.g. location/time/framing/gaze)

9
Encode statements

• Statement formally annotated
• ltsubjectgt ltmodifiergt ltpredicategt
• E.g. war best solution
• A thesaurus containing
• Terms for each part (155 in total)
• Relations between terms similar (72), opposite
  (108), generalization (10), specialization (10)
• E.g. war opposite diplomacy
• Relations in the thesaurus determine link type
  between statements

10
Automatic Linking Process

• STEP1 Using the thesaurus, generate related
  statements, by replacing iteratively terms
• E.g. from war best solution diplomacy best
  solution, war not solution
• STEP2 Query the repository to see whether the
  statement is present

11
Connect statements

• Create a graph of related statements
• Nodes are the statements (video segments), edges
  are either support or contradict

S8
S7
S0
S1
S6
S2
S9
S4
support
S3
S5
contradict
12
Author/Annotator support

• Capability of generating different arguments
  depends on the quality of the Semantic Graph
• Statements (and corresponding video segments) not
  connect are lost for generation
• Our case out of 118 statements 54 were not
  connected
• Measure the performance of the automatic linking
  process

13
Indices for statements

• Measure how many statements are generated from a
  given one
• depends on the quantity of the relations in the
  thesaurus
• Measure how many generates statements are present
  in the repository
• Depends on correctness of the relations in the
  thesaurus

14
Index for relations

• Measure how a particular relation in the
  thesaurus is performing
• If a generated statement is present in the
  repository, the relations used to generate it get
  one point on a hit score, otherwise one point on
  the miss score
• The ratio hit/miss gives an idea of the semantic
  accuracy of the relation with respect to the
  repository

15
Current/future Work

• Automatic Relation suggestion
• Start with a fully connected thesaurus, keep only
  best relations
• Suggest best relation to add to existing ones
• Linking Process tuning
• Currently 3 iterations for performance, but the
  process runs of-line more iterations possible
• Different repositories (VJ project)

16
Conclusions

• New documentary production mechanism, multiple
  versions
• Different authoring, author does not have full
  control anymore
• Authoring support needed

17
Questions?

• Thanks for your attention

18
Pointers Acknowledgments

• This presentation and Demo available at
• http//www.cwi.nl/media/demo/IWA/
• This research was funded by the Dutch national
  ToKeN2000 I2RP and CHIME projects.

19
Author/Annotator support

• Provide means to measure the performance of the
  creation of the Semantic Graph
• Reengineer the Semantic Graph generation
• Changing annotations
• Changing relations in the Thesaurus

20
What do you think of the war in Afghanistan?
I am not a fan of military actions
I cannot think of a more effective solution
War has never solved anything
Two billions dollar bombs on tents
21
Toulmin model
57 Claims, 16 Data, 4 Concessions, 3 Warrants, 1
Condition