IIT Bombay in ILILMT Consortium

1
IIT Bombay in IL-IL-MT Consortium

• Multiway lexicon, Lexical Disambiguation etc.
  Horizontals
• Marathi Vertical

2
Existing IITB Architecture for MT as specified in
the EoI
3
Experience gathered from existing work A
Multiway Lexicon Scenerio
shikshan
education (akoprocess)
education
shikkhaa
shikshaa
Disambiguating restriction
4
Lexical Disambiguation

• Given a text containing shikshaa in Hindi, one
  should hit education(akoprocess)
• Then pick up shikkhaa in Bengali and prashikshan
  in Marathi
• Disambiguation algorithms with language dependent
  and language independent components needed

5
Ancillary processes

• Root identification in the concerned language
  morphology analysis
• Morph generation after transformation rule
  application and lexical substitution
• Case marker/post position generation
• Syntax planning

6
Evaluation Metric Determination

• Automated as well as manual
• N-gram matching (BLEU/NIST like)
• Examine existing strategies IIITHs approaches
• IL-IL texts expected to be well aligned
• N-gram matching might work well

7
Existing Capability

• MCIT funded project on Technology Development in
  Indian Languages (first phase completed)
• World Bank funded Project on Development Gateway
  Foundation Language Technology Part
• MCIT funded Media Lab Asia Project Meaning
  Based, Multilingual Search Engine in the
  Agricultural Domain- AGroExplorer
• TCS funded project on Laboratory for Intelligent
  Internet Research

8
Relevant Language Resources and Tools
developed/under_development (1/2)

• Morph Analysers for Marathi and Hindi
• PoS taggers for Marathi and Hindi
• 3-way lexicon for Marathi, Hindi and English in
  the agricultural domain
• Concept dictionaries linked to Marathi, Hindi and
  English Words

9
Relevant Language Resources and Tools
developed/under_development (2/2)

• WordNets for Hindi and Marathi
• Limited Word sense disambiguator for Hindi using
  the Hindi WordNet
• Limited Semantics Extractor for English
• Syntax planners for Marathi and Hindi
• Morphers for Marathi and Hindi

10
Existing Team

• 5 PhD scholars
• 6 M.Tech students
• 8 B.Tech/B.E. students
• 3 linguists
• 7 lexicographers/lexiconists

11
Websites for Publications and Resources

• www.cfilt.iitb.ac.in
• www.cse.iitb.ac.in/pb
• www.cse.iitb.ac.in/laiir
• www.mlasia.iitb.ac.in
• (please also see the publications list in the
  Expression of Interest Document of 15th February,
  2006)