<s>
Hybrid	B-General_Concept
machine	I-General_Concept
translation	I-General_Concept
is	O
a	O
method	O
of	O
machine	B-Application
translation	I-Application
that	O
is	O
characterized	O
by	O
the	O
use	O
of	O
multiple	O
machine	B-Application
translation	I-Application
approaches	O
within	O
a	O
single	O
machine	B-Application
translation	I-Application
system	I-Application
.	O
</s>
<s>
The	O
motivation	O
for	O
developing	O
hybrid	B-General_Concept
machine	I-General_Concept
translation	I-General_Concept
systems	O
stems	O
from	O
the	O
failure	O
of	O
any	O
single	O
technique	O
to	O
achieve	O
a	O
satisfactory	O
level	O
of	O
accuracy	O
.	O
</s>
<s>
Many	O
hybrid	B-General_Concept
machine	B-Application
translation	I-Application
systems	I-Application
have	O
been	O
successful	O
in	O
improving	O
the	O
accuracy	O
of	O
the	O
translations	O
,	O
and	O
there	O
are	O
several	O
popular	O
machine	B-Application
translation	I-Application
systems	I-Application
which	O
employ	O
hybrid	O
methods	O
.	O
</s>
<s>
This	O
approach	O
to	O
hybrid	B-General_Concept
machine	I-General_Concept
translation	I-General_Concept
involves	O
running	O
multiple	O
machine	B-Application
translation	I-Application
systems	I-Application
in	O
parallel	O
.	O
</s>
<s>
Most	O
commonly	O
,	O
these	O
systems	O
use	O
statistical	B-General_Concept
and	O
rule-based	B-General_Concept
translation	O
subsystems	O
,	O
but	O
other	O
combinations	O
have	O
been	O
explored	O
.	O
</s>
<s>
For	O
example	O
,	O
researchers	O
at	O
Carnegie	O
Mellon	O
University	O
have	O
had	O
some	O
success	O
combining	O
example-based	B-General_Concept
,	O
transfer-based	B-General_Concept
,	O
knowledge-based	O
and	O
statistical	B-General_Concept
translation	I-General_Concept
sub-systems	O
into	O
one	O
machine	B-Application
translation	I-Application
system	I-Application
.	O
</s>
<s>
This	O
approach	O
involves	O
using	O
statistical	B-General_Concept
data	O
to	O
generate	O
lexical	B-Application
and	O
syntactic	O
rules	O
.	O
</s>
<s>
The	O
input	O
is	O
then	O
processed	O
with	O
these	O
rules	O
as	O
if	O
it	O
were	O
a	O
rule-based	B-General_Concept
translator	O
.	O
</s>
<s>
This	O
approach	O
still	O
suffers	O
from	O
many	O
problems	O
of	O
normal	O
statistical	B-General_Concept
machine	I-General_Concept
translation	I-General_Concept
,	O
namely	O
that	O
the	O
accuracy	O
of	O
the	O
translation	O
will	O
depend	O
heavily	O
on	O
the	O
similarity	O
of	O
the	O
input	O
text	O
to	O
the	O
text	O
of	O
the	O
training	O
corpus	O
.	O
</s>
<s>
As	O
a	O
result	O
,	O
this	O
technique	O
has	O
had	O
the	O
most	O
success	O
in	O
domain-specific	O
applications	O
,	O
and	O
has	O
the	O
same	O
difficulties	O
with	O
domain	O
adaptation	O
as	O
many	O
statistical	B-General_Concept
machine	I-General_Concept
translation	I-General_Concept
systems	O
.	O
</s>
<s>
The	O
most	O
common	O
technique	O
used	O
in	O
multi-pass	O
machine	B-Application
translation	I-Application
systems	I-Application
is	O
to	O
pre-process	B-Application
the	O
input	O
with	O
a	O
rule-based	B-General_Concept
machine	I-General_Concept
translation	I-General_Concept
system	O
.	O
</s>
<s>
The	O
output	O
of	O
the	O
rule-based	B-General_Concept
pre-processor	B-Application
is	O
passed	O
to	O
a	O
statistical	B-General_Concept
machine	I-General_Concept
translation	I-General_Concept
system	O
,	O
which	O
produces	O
the	O
final	O
output	O
.	O
</s>
<s>
This	O
technique	O
is	O
used	O
to	O
limit	O
the	O
amount	O
of	O
information	O
a	O
statistical	B-General_Concept
system	O
need	O
consider	O
,	O
significantly	O
reducing	O
the	O
processing	O
power	O
required	O
.	O
</s>
<s>
It	O
also	O
removes	O
the	O
need	O
for	O
the	O
rule-based	B-General_Concept
system	O
to	O
be	O
a	O
complete	O
translation	B-Application
system	I-Application
for	O
the	O
language	O
,	O
significantly	O
reducing	O
the	O
amount	O
of	O
human	O
effort	O
and	O
labor	O
necessary	O
to	O
build	O
the	O
system	O
.	O
</s>
