Building a virtual European Institute of Human Language Technologies

1
Building a virtual European Institute of Human
Language Technologies

• Steven Krauwer
• Utrecht University / ELSNET
• Steven.Krauwer_at_elsnet.org

2
Overview

• What is ELSNET
• What we do
• Instruments we use
• Our internal structure
• Some practical points
• The virtual European Institute

3
What is ELSNET

• European Network in Human Language Technologies
  (ca 140 academic and industrial member
  organisations)
• Funded by the European Commission
• Created in 1991 as one network out of
  (eventually) ca 25
• Objectives
• bringing together the language and speech
  communities
• bringing together academia and industry
• facilitating RD in language and speech
  technology
• Info elsnet_at_elsnet.org http//www.elsnet.org

4
Main areas of activity

• Main ingredients of European Networks of
  Excellence (from 1991 until the redefinition of
  the concept in 2003, see later)
• Research coordination
• Training
• Information dissemination
• Knowledge transfer between academia and industry

5
Research coordination

• Direct coordination very hard every research
  project has a funder who has his own expectations
  and claims
• Indirect coordination more promising
• Use funds to take new joint initiatives
• Use funds to bring together people who might
  benefit from each others activities
• Use funds to establish a longer term research
  agenda (roadmapping)

6
Training

• Addressing people already in the field or just
  entering
• Summer schools on typical ELSNET topics for those
  who have more time than money
• Bullet courses for those who have more money than
  time
• Mostly addressing academics
• Workshops or events on training related issues
• Tutorials on special ELSNET topics
• Curriculum development 1 year Masters curriculum
  in Language and Speech Technology

7
Information dissemination

• Website for internal communication within the
  community, but also to create awareness in the
  outside world
• Mailing lists (announcements, discussions, job
  ads)
• Paper magazine (proactive dissemination)
• Central information point

8
Knowledge transfer

• Important, but very hard to find the right model
• Best approximation
• Best practice projects
• Best practice website
• Best practice summer schools and courses
• Providing access to expertise (directories of
  experts and organisations world-wide)

9
The ELSNET temple
roadmapping
training
info dissemination
tech transfer
resources, standards, evaluation
10
Missions

• General research aims help increasing our
  knowledge and understanding of language and
  speech technology
• What our funders want improving our (Europes)
  competitive position
• Special language touch strong focus on
  resources, standards and evaluation
• Special European touch let RD for smaller
  languages benefit from what is happening for the
  major languages

11
Languages

• The Ethnologue (http//www.ethnologue.org)
• Europe 230 languages
• The Americas 1013 languages
• The Pacific 1311 languages
• Africa 2058 languages
• Asia 2197 languages

12
Our role

• Try to do things that would not have happened
  otherwise
• Try to act as a catalyst rather than as a funder
• Try to act as a bridge between communities

13
Measuring success

• Very hard, because you cannot prove that things
  would have been otherwise if it hadnt been for
  ELSNET
• Bureaucrats love objective measures such as
  participation in events, hits on the website,
  circulation of the newsletter, but I dont really
  believe in them as real success indicators

14
Some priorities

• Strong emphasis on generic issues with a
  (potentially) long term impact for the community
  at large, such as
• Standards
• Language Resources
• Evaluation
• Strong emphasis on interdisciplinary activities
  (in the ELSNET context mostly speech/language)
• Try to develop common visions of the future
  (roadmapping)

15
What is a roadmap

• A broadly supported vision of where our field is
  going (research, technology, market)
• Roadmapping as we see it is not about predicting
  the future but about managing expectations
• A coherent, consistent and broadly supported view
  should help us ( researchers, developers,
  providers, funders, educators) to
• identify main challenges
• set intermediate milestones
• concentrate efforts
• measure progress and (if necessary) adjust goals

16
Some instruments

• Paying for research not allowed (and would not
  have any added value anyway)
• Travel grants for staff and PhDs risky (no limit
  to what people will ask) but useful to make your
  community feel they benefit directly from it
• Funding invited speakers on specific ELSNET
  topics
• Organizing workshops, panels, etc to bring people
  together or to promote the goals (awareness)

17
Some more instruments

• Either partial reimbursement or one-for-one
• Organise events in conjunction with main
  conferences
• Try to convert requests for funding into
  co-organization of activities following from our
  programme of work
• Publish books on the basis of summer schools to
  reach a broader audience
• Endorsement of events organised by members or of
  other high quality events

18
Typical ELSNET topics

• Illustrated by Summer School topics
• 1993 Prosody
• 1994 Corpus based methods
• 1995 Spoken dialogue systems
• 1996 Multilinguality
• 1997 Robustness

• 1998 Multimodality
• 1999 The lexicon
• 2000 Access to information
• 2001 Annotation of corpora
• 2002 Evaluation
• 2003 Computer assisted language learning

19
Industrial involvement

• 60 of our members are academic institutes, 40
  are industrial
• Industrials hard to mobilize because they are
  driven by their own priorities
• Rather successful model let academics do all the
  work and invite industrials to join an industrial
  advisory panel (with early access to
  information), but without any commitment to do
  anything

20
Pragmatic points

• For all practical purposes we have adopted the
  hub and spoke model strong coordination point
  (the ELSNET Office) rather than a 100
  distributed approach
• Dedicated staff (e.g. 40 coordinator, 40 info
  collection and dissemination, 80 administrator)
  needed to keep things moving
• One single contact point for the network
  www.elsnet.org for the web, elsnet_at_elsnet.org for
  email

21
Decision taking

• Executive Board, ca 9 high level experts, meeting
  3 times per year (could be electronically if
  necessary)
• Cooptation in order to ensure proper thematic and
  geographical balance
• Task groups or committees for specific tasks
• Watch out for the celebrity syndrome!

22
Financial matters

• Average annual budget ca 300 keuro
• Keep the money in one place, if possible, in
  order to prevent fragmentation of the budget
• Reimbursement system
• Allocate funds to activities rather than to
  participants (allows for flexible task allocation
  and for joining in later of new participants)
• Allow for flexibility in order to be able to
  adapt to the dynamics of the field (emerging
  trends or needs)

23
Where does the money go

• Main cost items
• Coordination point (labour, ca 150 keuro/yr)
• Travel (moving around members plus invited
  experts)
• ELSNews (quarterly paper magazine, ca 60
  keuro/yr)
• Summer school (ca 20 keuro/yr)
• Forbidden items research, student grants,
  support for non-EU activities
• Lots of silly bureaucratic constraints

24
Other issues

• Internationalization Try to embed your
  activities internationally but check with your
  funders
• Paper magazine (quarterly) widely appreciated,
  more effective than electronic variants, but
  expensive (60 keuro pr year)
• Moving towards more permanent structures in
  Europe is hard
• EU is opposed to permanent funding
• National funders and industry not interested in
  supporting facilitation of European RD at large

25
Continuity

• ELSNET has always presented itself as a permanent
  structure, but technically it is a series of
  independent contracts, with no EC commitment
  beyond duration of the contract
• We have set up a (light) legal entity, the ELSNET
  Foundation to own the brand, to ensure some
  continuity and to be able to enter into contracts
  (e.g. publishers)

26
Our immediate future

• Present funding contract expires this summer, no
  follow-up funding yet
• Language and speech technology have disappeared
  from the EU research agenda for the new Framework
  Programme (FP6)
• Replaced by Interfaces, Cognitive systems,
  Knowledge management
• Not the first funding gap, but the conditions are
  very bad we can still stretch it until Summer
  2005 if we keep expenditure low
• But FP7 is already in the oven, and seems to be
  more promising

27
The new network concept

• EU has redefined the Network of Excellence
  concept
• Moving away from spreading excellence to
  boosting excellence, resulting in an elitist
  approach
• Real research coordination much more prominent
• Partners are obliged to commit part of their
  research resources
• Support for a limited period, result should be a
  self-sustaining permanent research coordination
  framework

28
The Virtual Institute

• Still a dream the creation of a virtual European
  institute in language and speech technology
• Should be part of the European Research Area
• First attempt failed Why this complex structure
  on top a an otherwise successful network?

29
End of this presentation

• More info
• http//www.elsnet.org
• Steven.Krauwer_at_elsnet.org