Searching Indexed Bilingual Knowledge Banks

1
Searching Indexed Bilingual Knowledge Banks

• by Kee Tuan Hwa

2
Outline

• Introduction
• Proposed Method
• Summary

3
Introduction

• Types of machine translation
• - Rules-based machine translation
• - Knowledge-based machine translation
• - Statistical-based machine translation
• - Example-based machine translation
• (EBMT)

4
EBMT Architecture

• Source

• Find the best matching in the bilingual corpus

5
EBMT Architecture (cont.)
6
EBMT Architecture (cont.)
7
EBMT Architecture (cont.)
Recombination
8
Examples of EBMT

• Gaijin System - Uses a bilingual lexicon and
  transfer rules
• MSR-MT - Uses MindNet, logical form

9
Weaknesses
10
Weaknesses

• a finite means for generating the potential
  infinity of linguistic forms a speaker-hearer can
  produce or recognize (Chomsky,1928)

11
Weaknesses
30k of S-SSTC
S-SSTC

• The size of BKB is big, it will take time to
  perform searching. The graph complexity is
  exponential, O(n)en

S-SSTC
S-SSTC
S-SSTC
S-SSTC
S-SSTC
S-SSTC
S-SSTC
S-SSTC
S-SSTC
S-SSTC
12
Weaknesses

• Poor classification of S-SSTC in the Bilingual
  knowledge Bank (BKB)

S-SSTC
S-SSTC
S-SSTC
S-SSTC
S-SSTC
S-SSTC
S-SSTC
S-SSTC
S-SSTC
13
Goal and Objective

• Classification of STREE and SNODE correspondence
  in BKB for effective retrieval and translation

14
Proposed System Design
15
Index Construction
16
Indexed BKB
Indexed S-SSTC
Indexed BKB
Indexed S-SSTC
Indexed S-SSTC
Indexed S-SSTC
Indexed S-SSTC
Indexed S-SSTC
17
Pivot BKB

• Pivot BKB
• - clustering the S-SSTC
• - categorize the SNODE, STREE based
• on pattern-based POS
• - use the modification of inverted file
• indexing

18
Pivot BKB (cont.)
IP seeV theDet bigAdj bird N ??sayaP
lihatV burungN besarAdj ituDet
SNODE CORRESPONDENCE 1.5.1 I P ??saya P
1.5.2 see V ??lihatV 1.5.3 the Det
??itu Det 1.5.4 big Adj
??besarAdj 1.5.5 bird N ??burungN
Stree CORRESPONDENCE 1.6.1 P V Det Adj
N ?? P V N Adj Det
I see the big bird ??saya lihat burung
besar itu 1.6.2 V Det Adj N ?? V
N Adj Det see the big
bird ?? lihat burung besar itu 1.6.3 Det Adj
N ?? N Adj Det the big
bird ?? burung besar itu
Name of Indexed BKB a.xml
19
Pivot BKB (cont.)
Name of Indexed BKB
Sense of SNODE
ltPgt
ltVgt
- I ,lta.xml1.5.12gt
- see ,lta.xml1.5.22gt
Index of SNODE
ltDetgt
ltAdjgt
- the ,lta.xml1.5.31gt
- big ,lta.xml1.5.44gt
ltNgt
Name of Indexed BKB
- bird ,lta.xml1.5.53gt
ltP V Det Adj Ngt
ltV Det Adj Ngt
- I see the big bird ,lta.xml1.6.1gt
- see the big bird ,lta.xml1.6.2gt
Index of STREE
ltDet Adj Ngt
- the big bird ,lta.xml1.6.3gt
20
Pivot BKB (cont.)
theDet oldAdj manN walked V ?? orangN
tuaAdj iniDet berjalanV
SNODE CORRESPONDENCE 2.5.1 theDet ?? ini Det
2.5.2 oldAdj ?? tua Adj 2.5.3 manN ??
orang N 2.5.4 walked V ?? berjalan V
Stree CORRESPONDENCE 2.6.1 Det Adj N V
?? N Adj Det V the old man
walked ?? orang tua ini berjalan 2.6.2 Det
Adj N ?? N Adj Det the old
man ?? orang tua ini
Name of Indexed BKB a.xml
21
Pivot BKB (cont.)
ltPgt
ltVgt
- I ,lta.xml1.5.12gt
- see ,lta.xml1.5.22gt
- walked ,lta.xml2.5.41gt
ltDetgt
ltAdjgt
- the ,lta.xml1.5.31gt
- big ,lta.xml1.5.44gt
- the ,lta.xml2.5.12gt
- old ,lta.xml2.5.21gt
ltNgt
- bird ,lta.xml1.5.53gt
- man ,lta.xml2.5.32gt
ltP V Det Adj Ngt
ltV Det Adj Ngt
- I see the big bird ,lta.xml1.6.1gt
- see the big bird ,lta.xml1.6.2gt
ltDet Adj Ngt
ltDet Adj N Vgt
- the big bird ,lta.xml1.6.3gt
- the old man walked ,lta.xml2.6.1gt
- the old man ,lta.xml2.6.2gt
22
Summary

• size(n)?1/speed(n)
• Classification and indexing to perform effective
  retrieval and translation