Language Technology in European Funding Programmes

1
Language Technology in European Funding Programmes

• Kimmo RossiDG Information society and MediaUnit
  E1 Language Technology, Machine Translation

2
DG INFSO E in Luxembourg
3
Introducing DG INFSO

• DG Information Society and Media
• 10 directorates
• not only (research) funding agency policy
  making, market regulation
• strategic policy framework i2010
• most sizeable implementation task ICT programme
  of FP7 (9.1 B)

4
Directorate E Digital Content cognitive
systems

• located in Luxembourg
• 7 units, 130 people
• Main themes
• Digital libraries
• Public sector information
• Language technology
• Cognitive systems, robotics
• Safer Internet
• content production, (re-)use

5
Unit E1 Language Technology, Machine Translation

• Until 30/6/2008 Interaction Interfaces
• New name implies more emphasis on (novel)
  language technology
• Changed political, technologic and societal
  context
• EU of 23 official languages
• Breakthrough of data-driven MT
• Community-based approach to content production
  and use (Web 2.0)

6
Brief history (1993-2003)

• MLIS-FP3-FP4-LE-ESPRIT
• Dedicated Language Technology projects (Machine
  Translation, translation tools, terminology,
  standardisation, LRs)
• some important seed technology (e.g. Translators
  Workbench, EBMT)
• About 100 small projects (less than 1 MEUR)
• FP5 (5th framework programme)
• Applications of Language Technology (cars, mobile
  services, information retrieval, knowledge mgmt)
• 90 projects, 130 MEUR funding
• Mid-sized projects 0.7 2.5 MEUR

7
From history to present

• FP6 (6th framework programme)
• 28 (20 with LT) projects, 120 MEUR
• Larger projects (up to 15 MEUR)
• Non-linguistic interaction about 12 projects
  TnD, TAI-CHI, ARTTS, SATIN,
• Pure language technology TC-STAR, TALK,
  EUROMATRIX, SMART, LUNA,
• Multimodal interaction CHIL, AMI, AMIDA,
• FP7 (7th framework programme)
• Challenge 2 cognitive systems, robotics,
  interaction
• Objective 2.2 language-based interaction

8
Current projects

• EuroMatrix (www.euromatrix.net )
• Intoducing linguistics into SMT
• Factored SMT (treelet alignment, morphology)
• Improvement by various means (combining engines
  etc.)
• Inventory of MT systems and language resources
  for SMT
• SMART (www.smart-project.eu )
• Optimising SMT (algorithms, parameters, combining
  SMT engines)
• Machine learning techniques (e.g. kernel methods)
• towards adaptive MT
• ITALK (italkproject.org )
• iCub robot learns language by doing
• integrates action, social and linguistic skills,
  cognitive development
• language learning from scratch, for simple tasks
• ROSSI (www.rossiproject.net )
• how language development is linked to physical
  experience
• sensori-motor grounding of human
  conceptualisation and language use
• neurologic basis underlying the use of verbs and
  nouns
• grounding of object affordances

9
Current projects

• ALEAR (www.alear.eu )
• language evolution in robot populations
• baseline language games (Luc Steels),
  recruitment theory
• self-organisation of conceptual frameworks and
  communication systems
• POETICON (www.poeticon.eu )
• linking sensori-motor representations with
  linguistic ones
• extending Lexicon into Praxicon (grammar and
  lexicon of action)
• EMIME (www.emime.org )
• unified modelling of speech recognition and
  synthesis
• personalised speech synthesis (your voice
  speaking chinese)
• reversibility of statistical modelling techniques
• FlareNet (http//www.ilc.cnr.it/flarenet/ )
• thematic network on Language Resources
• what are Language Resources?
• promotes interaction of LR stakeholders
• inventory and roadmap for action
• open for new members join now!

10
Trends

• New requirements new approaches
• From Web 1.X to Web 2.0 we are all content
  producers
• From static and uni-directional to dynamic,
  volatile, interactive, collaborative
• Translations are needed on the fly
• Are language technologies up to the task?
• What happens to online content
• Disappearing document
• What is on the Internet? Who knows? Google?
• Europa web site 6 million documents
• Disappearing distinction between content and
  service
• How to manage (automatically?) the multilingual
  online content
• From service to self-service
• Travelling, banking, house-buying
• Need for language-literate systems
• Multilingualism on the rise
• In the EU (from 4 to 23 languages) and the
  global dimension

11
Challenges

• Machine translation new paradigms
• adaptive, self-learning MT systems
• MT that learns from its mistakes
• learning through understanding and vice versa
• Is there any future for RBMT?
• Bringing scientific communities together
• learning a common language among researchers
• detecting common interest mutual benefit
• SCs can learn from each other
• Language resources
• Exploit the hidden treasures (e.g. public sector
  resources)
• Improve usability of existing resources
• Identify and address gaps in coverage
• Reusable, standardized, automated collection
  (e.g. from Web)
• Towards automation harvesting LRs from the Web

12
Funding opportunities ahead?

• Language-based interaction priorities will be
  defined in next FP7 Work programme (covering call
  4)
• http//cordis.europa.eu/fp7/dc
• Online multilingualism creation of a single
  European information space. Watch out for future
  actions in the ICT-PSP programme
• http//ec.europa.eu/information_society/activitie
  s/ict_psp

13
Thank You for your attention

• Contact
• Kimmo.Rossi_at_ec.europa.eu