Machine Translation from English to HindiMarathi - PowerPoint PPT Presentation
1 / 8
Title:
Machine Translation from English to HindiMarathi
Description:
What is the expected category of contribution: Framework for translation from ... going Masters work in idiom discovery, clause boundary detection, abbreviation ... – PowerPoint PPT presentation
Number of Views:1634
Avg rating:3.0/5.0
Title: Machine Translation from English to HindiMarathi
1
Machine Translation from English to Hindi/Marathi
- Proposers Jayprasad J. Hegde and Ananthakrishnan
Ramanathan - jjhegde,anand_at_cdacmumb
ai.in - Team headed by Dr. M. Sasikumar
- Team Size 7
- Name of the company CDAC Mumbai (formerly NCST)
- http//www.cdacmumbai.in/
- Language/Language pair English to Hindi/Marathi.
- What is the expected category of contribution
Framework for translation from English to
languages in the North-Western belt (Hindi,
Marathi, Gujarati, etc.) Hindi generation,
Transliteration components
2
Strength of the organization
- Technical Capabilities
- Strong RD background in areas such as AI, NLP,
IR, Graphics, Networking, Databases, and Software
Engineering - RD Budget
- 11 Crores per annum
- Decision Process
- Through committees and approval of ED and DG
- Manpower Involved
- 150 regular scientists 90 project engineers
- Some previous collaborations with
universities/RD institutions - Fraunhoffer Institute, Germany,
- University of Pennsylvania,
- Indian Institute of Technology, Bombay,
- Concordia University, Canada, etc.
- SNDT
- Goa University
3
Current and Previous Work
- MaTra2 An English-Hindi Fully-automatic Machine
Translation System (on-going) - MaTra A Human-aided Machine Translation System
from English to Hindi - Setu A Hindi-English Cross Language Information
Retrieval System (on-going) - Completed Masters work in Transliteration
(English-Hindi, Hindi-English) and EBMT - On-going Doctoral work in Statistical MT
- On-going Masters work in idiom discovery, clause
boundary detection, abbreviation detection, and
translation memory
4
Previous Work (contd.)
- Older Projects
- (not in chronological order)
- Vaakya
- Khoj An IR system
- narch An news archive for PTI allowing access to
news snippets
5
Architecture of existing/proposed MT system
6
Goals of our MT system
- Understandable translations for wide-coverage
rather than perfect translations for a small
range (domain) - Work well for incomplete and ill-formed sentences
- Example application To provide indicative
translation for documents on the Web
7
Select features of the existing system
- Structural and Lexical Transfer
- Rule-bases for Hindi generation
- Target language independent intermediate
structured representation - Machine learning technique for transliteration
- Transliteration (English Hindi, existing 80
accuracy) - Tested approach for Hindi
- 60 acceptability on unseen sentences from
general news corpus capturing a wide variety of
sentences
8
Summary
- General Framework for MT. To be demonstrated for
a language in the north-western belt Marathi. - Ongoing work in focused problems in MT
- Transliteration
- IL generation
Recommended
CrystalGraphics Presentations
