SMART Kickoff Meeting Welcome and Introduction

1
SMART Kick-off MeetingWelcome and Introduction

• Nicola Cancedda
• October 23, 2006

2
Motivation

• More than half of the citizens of the EU do not
  speak a second language
• Expanding global demand of tools for automatic
  translation and cross-language retrieval,
  clustering and categorization
• Statistical approaches mainstream in research,
  but still suffering from shortcomings, preventing
  their diffusion
• Many recent advances in ML are only starting to
  hit the field

SMART is an attempt to solve identified
shortcomings of MT and CLTIA with the methods of
Statistical Learning
3
The SMART project

• Information Society Technologies Programme
• Sixth Framework Programme, Specific Target
  Research Project (STReP)
• Start date October 1, 2006
• Duration 3 years

4
The SMART Consortium
5
The SMART Consortium
6
Work-packages

• Scientific workpackages
• WP2 Advanced Statistical Translation Models
• WP3 Advanced Language Models
• WP4 Model Adaptation and Combination
• WP5 Cross-Language Textual Information Access
• WP6 Evaluation
• WP7 Dissemination and Exploitation
• WP1 Project Management and Co-ordination

7
WP2 Advanced Statistical Translation Models

• Going beyond phrase-based log-linear models in
  SMT
• T2.1 Characterization of current (SMT) approaches
• M1-12. Deliverable 2.1
• T2.2 Theoretical models of Markov-based
  approaches
• M1-12. Deliverable 2.1
• T2.3 Application of Markov Approaches to SMT
• M10-24. Deliverable 2.2
• T2.4 Efficient Implementation of Developed
  Methods
• M13-33. Deliverables 2.3, 2.4, 2.5.

8
WP3 Advanced Language Models

• Better LMs for improved translation fluency and
  CLTIA
• T3.1 Kernels for Language Models
• M1-12. Deliverable 3.1.
• T3.2 Learning Discriminative Language Models
• M7-21. Deliverable 3.2.
• T3.3 Integration with Translation Model
• M16-33. Deliverables 2.3, 5.3, 2.4, 5.4.

9
WP4 Model Adaptation and Combination

• Going beyond batch training to cope with data
  scarcity and leverage user feedback
• T4.1 Mixture-based Approaches for the Adaptation
  of Translation Models
• M1-33. Deliverables 4.1, 2.3, 5.3, 2.4, 5.4.
• T4.2 On-line Training Algorithms for Incremental
  Training of Translation Models
• M1-33. Deliverables 4.1, 2.3, 5.3, 2.4, 5.4.
• T4.3 Algorithms for achieving Domain and Genre
  Independence via Model Combination
• M1-33. Deliverables 4.1, 2.3, 5.3, 2.4, 5.4.
• T4.4 On-line Learning Algorithms for
  Computer-Aided Translation
• M1-33. Deliverables 4.2, 2.3, 5.3, 2.4, 5.4.

10
WP5 Cross-Language Textual Information Access

• Going beyond single-word translation and
  leveraging synergies with latent-semantics
  approaches
• T5.1 Multilingual Lexicon Extraction
• M1-33. Deliverables 5.1, 5.3, 5.4.
• T5.2 Latent Language-Independent Analysis
• M4-33. Deliverables 5.2, 5.3, 5.4.
• T5.3 Model Adaptation and On-line Aspects of
  CLTIA
• M7-33. Deliverable 5.4.

11
WP6 Evaluation

• In-vitro and in-vivo testing
• T6.1 Benchmarking of Machine Translation
  Technology
• M1-30. Deliverable 6.2
• T6.2 User Evaluation of Adaptive MT Technology
• M1-12, 24-27, 33-36. Deliverables 6.1, 6.3.
• T6.3 Benchmarking of Cross-Language Textual
  Information Access
• M1-30. Deliverable 6.2.
• T6.4 User Evaluation of CLIR comprehension aids
• M1-12, 24-27, 33-36. Deliverables 6.1, 6.3.

12
WP7 Dissemination and Exploitation

• Making the best of it
• T7.1 Dissemination
• M1-36. Website. Scientific workshop.
• T7.2 Technology Watch
• M7-36. Techwatch (private) portion of the website
• T7.3 Exploitation Planning and IP Valorization
• M1-36. Deliverable 7.2.
• T7.4 Technology Demonstration and Showcasing
• M25-36. Deliverable 7.1. Business workshop.

13
Expected results and their use

• Scientific results
• Radically new research directions in SMT and
  CLTIA
• Incremental improvements in existing SMT and
  CLTIA systems
• Improvements in statistical learning methods by
  facing new challenges in hard real-world tasks
• Building conceptual bridges between all
  involved communities
• Dissemination
• Scientific workshop (See Task 7.1)
• Presentations to conferences and workshops
  (including relevant evaluation events)
• Journal articles
• Scientific portion of the website

14
Expected results and their use

• User results
• More usable SMT systems
• Performing better
• Adapting to new domains and over time
• Producing more fluent output
• Improved access to multilingual document
  collections
• CLIR potential eventually expressed via coupling
  with comprehension aids
• Exploitation routes
• Internal, through partners business units
• External, through technology licensing
• Non-scientist-oriented section of the project
  website
• Business-oriented workshop at the end of the
  project (See Task 7.4)
• Possibly publicly available showcase on the
  WikiPedia scenario