Multilingual Information Access in a Digital Library

1
Multilingual Information Access in a Digital
Library

• Vamshi Ambati, Rohini U, Pramod, N Balakrishnan
  and Raj Reddy
• International Institute of Information Technology
• Hyderabad, India

2
Context

• Digital Library of India
• 155,000 English books
• 145,000 Other language books
• Population of literates
• 20 of India understand English
• 80 can not

3
Multilingual Access to Information

• Retrieve a book
• By metadata
• By keyword / content
• Cross Lingual Information Retrieval
• Read a book
• Help understand sentences in a language
• Help understand sentences across languages
• Machine Translation

4
Approaches to Multilingual Access

• Cross Lingual Retrieval
• Translate Query to Document Language
• Translate Document to Query Language
• Machine Translation
• Knowledge Based Approaches
• Corpus Based Approaches
• Hybrid Approaches

5
Challenges in Multilingual Access

• Corpus Based Approaches
• Unavailability of Parallel Corpus for pairs of
  languages
• Unavailability of Computational Linguistics
  Resources
• Dictionary Based Approaches
• Unavailability of multiple bilingual dictionaries

6
Resources

• Universal Dictionary
• Conceived and implemented by Michael Shamos at
  CMU, USA
• ITRANS
• A transcription scheme and associated tool built
  by IISc, IIIT and CMU
• Corpus
• Data Entry by TTD and DLI project
• TIDES project

7
Universal Dictionary
8
How are we doing it

• Cross Lingual Search (Identify Information)
• Dictionary lookup
• User feedback based
• Lucene Search Engine
• Machine Translation (Understand Information)
• Corpus based technique (EBMT)
• Dictionary based word-word lookup
• Good-enough translation vs Perfect translation

9
Cross Lingual Retrieval
10
Cross Lingual Retrieval
11
Reading Assistant System
12
Reading Assistant
13
Status Today

• CLIR for 6 languages
• MT for 3 languages
• Shakti (a knowledge based MT system)
• Parallel Corpus for Hindi-Eng
• UDICT
• About 40 Foreign Languages
• 6 Indian Languages

14
What more is needed?

• UDICT
• Improving coverage of existing languages
• Adding new languages
• Machine Translation
• Corpus acquisition
• State of art techniques applied to Indian
  Languages
• Multi-way parallel corpus development
• Textual format for the books
• Books currently are in Image formats
• OCR should be developed for textual content

15
Thank You

• Questions ?