The Challenge of

1
Your photo here
The Challenge of Syntax-Based Machine Translation
Ondrej Bojar Computational Linguistics
ÚFAL, Charles University, Prague (visiting LT
Group, CSSE, Melbourne University)
2
Syntax-Based Machine Translation
Prekládají
lidé
slovo
od
slova
?
No
Do-translate
people
word
by
word
?
prekládat
translate
Translate
Maybe
lidé
struktura
people
structure
Analyze
Generate
Prekládají lidé strukturu?
Do people translate structure?
3
Translating Trees
prekládat
prekládat
Decompose
lidé
struktura
lidé
struktura
translate
Combine
translate
people
structure
people
structure
4
Results Discussion

• Tree translation performs poorly
• More steps ? More errors.
• Complex structure ? More conflicts.
• Future experiments
• Use smaller bits (decompose nodes).
• Improve components of the system.
• Lazy alternative given enough data

Maybe
Prekládají lidé
rovnou celé fráze
?
Do people translate
longer phrases at once
?