Example Based Machine Translation - PowerPoint PPT Presentation

1 / 21
About This Presentation
Title:

Example Based Machine Translation

Description:

Example Based Machine Translation. A Bilingual Dictionary was used for this Purpose. ... Example Based Machine Translation. More than around 400 rules added in ... – PowerPoint PPT presentation

Number of Views:628
Avg rating:3.0/5.0
Slides: 22
Provided by: tdilM
Category:

less

Transcript and Presenter's Notes

Title: Example Based Machine Translation


1
Example Based Machine Translation
Indian Institute of Science
2
Example Based Machine Translation
  • A Bilingual Dictionary was used for this
    Purpose.
  • The Bilingual dictionary is a set of 3
    dictionaries comprising of
  • Sentence Dictionary
  • Phrases Dictionary
  • Words Dictionary

3
Example Based Machine Translation
Ebmt Flow Diagram
2. Socket Connections
1.Request
Java Servlet Apache Tomcat
Client Browser
Machine Translation Programs running as
servers. One server for one Translation system.
4.Response
3. Result of Processing
4
Example Based Machine Translation
Hindi Corpora
  • Sentence Dictionary - 1900 Sentences
  • Phrases Dictionary - 35000 Phrases
  • Word Dictionary - 34000 Words

5
Example Based Machine Translation
Tamil Corpora
  • Sentence Dictionary - 40000 Sentences
  • Phrases Dictionary - 28490 Phrases
  • Word Dictionary - 52400 Words

6
Example Based Machine Translation
Kannada Corpora
  • Sentence Dictionary - 40000 Sentences
  • Phrases Dictionary - 12480 Phrases
  • Word Dictionary - 25000 Words

7
Example Based Machine Translation
  • More than around 400 rules added in Hindi, tamil
    and Kannada
  • Performance evaluated for good- enough
    translation

8
Example Based Machine Translation
Machine Evaluation
Aim To estimate the improvement in
performance of EBMT system due to addition of
rules and increasing the size of the corpora
The Methodology BLEU BiLingual Evaluation
Understudy, from IBM Research Lab
9
BLEU Score Results -Tamil
Where, 1 ? Performance of EBMT system without
Rules 2 ? Performance of EBMT system with 500
rules
10
BLEU Score Results Tamil (Contd)
Where, 1 ? Performance of EBMT system without
Rules 2 ? Performance of EBMT system with 500
rules
11
BLEU Score Results - Kannada
Where, 1 ? Performance of EBMT system without
Rules 2 ? Performance of EBMT system with 452
rules
12
BLEU Score Results Kannada(Contd)
Where, 1 ? Performance of EBMT system without
Rules 2 ? Performance of EBMT system with 452
rules
13
Manual Evaluation
Aim To demonstrate Human
assisted good enough Translation
  • The Methodology
  • A test bed of sentences along with its
    translation, which is done by machine is given to
    the user. For each sentence, the user will give
    ranking in the following range
  • Useless
  • Poor
  • Mediocre
  • Acceptable
  • Human Quality

14
Example Based Machine Translation
Manual Evaluation (Contd)
The Need Even though machines are evaluating
Machine Translation at large extent they give
credits only based on rules and their evaluation
is not up to the mark because
Machines cant understand context
Machines cant tolerate typographical errors
15
Manual Evaluation Screen Shot I
16
Manual Evaluation Screen Shot II
17
Manual Evaluation Results
18
EBMT Result 1
19
Ebmt -Result 2
20
Ebmt - Result 3
21
EBMT- Our Proposal
  • It has been planned to increase the words,
    phrases and sentences data base from its present
    level by adding 25,000 more words, phrases and
    sentences in Tamil and Kannada.
  • To write programmes to mine the web and to deduce
    rules from the parallel corpora.
  • To design an interface for the users to test out
    the machine translation system and augment it
    through suggestions using the Wikipedia approach.
  • CLICK HERE TO VIEW THE URL
Write a Comment
User Comments (0)
About PowerShow.com