<s>
METEOR	B-General_Concept
(	O
Metric	O
for	O
Evaluation	O
of	O
Translation	O
with	O
Explicit	O
ORdering	O
)	O
is	O
a	O
metric	O
for	O
the	O
evaluation	B-General_Concept
of	I-General_Concept
machine	I-General_Concept
translation	I-General_Concept
output	I-General_Concept
.	O
</s>
<s>
The	O
metric	O
is	O
based	O
on	O
the	O
harmonic	O
mean	O
of	O
unigram	B-Language
precision	O
and	O
recall	O
,	O
with	O
recall	O
weighted	O
higher	O
than	O
precision	O
.	O
</s>
<s>
It	O
also	O
has	O
several	O
features	O
that	O
are	O
not	O
found	O
in	O
other	O
metrics	O
,	O
such	O
as	O
stemming	B-General_Concept
and	O
synonymy	O
matching	O
,	O
along	O
with	O
the	O
standard	O
exact	O
word	O
matching	O
.	O
</s>
<s>
The	O
alignment	O
is	O
a	O
set	O
of	O
mappings	O
between	O
unigrams	B-Language
.	O
</s>
<s>
A	O
mapping	O
can	O
be	O
thought	O
of	O
as	O
a	O
line	O
between	O
a	O
unigram	B-Language
in	O
one	O
string	O
,	O
and	O
a	O
unigram	B-Language
in	O
another	O
string	O
.	O
</s>
<s>
The	O
constraints	O
are	O
as	O
follows	O
;	O
every	O
unigram	B-Language
in	O
the	O
candidate	O
translation	O
must	O
map	O
to	O
zero	O
or	O
one	O
unigram	B-Language
in	O
the	O
reference	O
.	O
</s>
<s>
Stages	O
are	O
run	O
consecutively	O
and	O
each	O
stage	O
only	O
adds	O
to	O
the	O
alignment	O
those	O
unigrams	B-Language
which	O
have	O
not	O
been	O
matched	O
in	O
previous	O
stages	O
.	O
</s>
<s>
Once	O
the	O
final	O
alignment	O
is	O
computed	O
,	O
the	O
score	O
is	O
computed	O
as	O
follows	O
:	O
Unigram	B-Language
precision	O
is	O
calculated	O
as	O
:	O
</s>
<s>
Where	O
is	O
the	O
number	O
of	O
unigrams	B-Language
in	O
the	O
candidate	O
translation	O
that	O
are	O
also	O
found	O
in	O
the	O
reference	O
translation	O
,	O
and	O
is	O
the	O
number	O
of	O
unigrams	B-Language
in	O
the	O
candidate	O
translation	O
.	O
</s>
<s>
Unigram	B-Language
recall	O
is	O
computed	O
as	O
:	O
</s>
<s>
Where	O
is	O
as	O
above	O
,	O
and	O
is	O
the	O
number	O
of	O
unigrams	B-Language
in	O
the	O
reference	O
translation	O
.	O
</s>
<s>
In	O
order	O
to	O
take	O
these	O
into	O
account	O
,	O
longer	O
n-gram	B-Language
matches	O
are	O
used	O
to	O
compute	O
a	O
penalty	O
for	O
the	O
alignment	O
.	O
</s>
<s>
In	O
order	O
to	O
compute	O
this	O
penalty	O
,	O
unigrams	B-Language
are	O
grouped	O
into	O
the	O
fewest	O
possible	O
chunks	O
,	O
where	O
a	O
chunk	O
is	O
defined	O
as	O
a	O
set	O
of	O
unigrams	B-Language
that	O
are	O
adjacent	O
in	O
the	O
hypothesis	O
and	O
in	O
the	O
reference	O
.	O
</s>
<s>
Where	O
c	O
is	O
the	O
number	O
of	O
chunks	O
,	O
and	O
is	O
the	O
number	O
of	O
unigrams	B-Language
that	O
have	O
been	O
mapped	O
.	O
</s>
