I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
14
, N
o.
2
,
A
pr
il
20
2
5
, pp.
1587
~
1596
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
2
.pp
1587
-
1596
1587
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
A
b
st
r
ac
t
i
ve
su
m
m
ar
i
z
at
i
on
u
si
n
g
m
u
l
t
i
l
i
n
gu
a
l
t
e
xt
-
to
-
t
e
xt
t
r
an
sf
e
r
t
r
an
sf
or
m
e
r
f
or
t
h
e
T
u
r
k
i
sh
t
e
xt
N
e
d
a A
li
p
ou
r
1
, S
e
r
d
ar
A
yd
ın
2
1
D
e
pa
r
t
m
e
nt
o
f
M
a
na
g
e
m
e
nt
I
n
f
o
r
m
a
t
i
on
S
ys
t
e
m
s
,
F
a
c
ul
t
y
o
f
E
c
on
om
i
c
s
a
nd
A
d
m
i
n
i
s
t
r
a
t
i
ve
S
c
i
e
nc
e
,
A
t
a
t
ü
r
k U
ni
ve
r
s
i
t
y,
E
r
z
u
r
um
,
T
ür
ki
ye
2
D
e
pa
r
t
m
e
nt
of
S
of
t
w
a
r
e
E
ngi
ne
e
r
i
ng,
F
a
c
ul
t
y of
E
ngi
ne
e
r
i
ng, A
t
a
t
ü
r
k U
ni
ve
r
s
i
t
y, E
r
z
ur
um
,
T
ür
ki
ye
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
O
c
t
22, 2023
R
e
vi
s
e
d
N
ov 17, 2024
A
c
c
e
pt
e
d
N
ov 24, 2024
Today, with the increase
in text
data, the application
of automatic
tech
niques
such
as
automatic
text
summarization,
which
is
one
of
the
most
critical
natural
language
processing
(NLP)
tasks,
has
attracted
even
more
att
ention
and
led
to
more
research
in
this
area.
Nowadays,
with
the
developm
ents
in
deep
learning,
pre
-
trained
sequence
-
to
-
sequence
(
text
-
to
-
text
t
ransfer
converter
(T5)
and
bidirectional
encoder
representations
from
transf
ormers
(BERT)
algorithm
)
encoder
-
decoder
models
are
used
to
obtain
th
e
mo
st
advanced
results.
However,
most
of
the
studies
were
done
in
the
Englis
h
language.
With
the
help
of
the
recently
emerging
m
onolingual
BERT
model
and
multilingual
pre
-
trained
sequ
ence
-
to
-
sequence
models,
it
has
led
to
the
use
of
state
-
of
-
the
-
art
models
in
languages
with
fewer
resources
and
s
tudies,
such
as
Turkish.
This
article
used
two
datasets
for
Turkis
h
text
summarization.
First,
Google
multilingual
t
ext
-
to
-
text
transfer
trans
former
(
mT5
)
-
small
model
was
applied
on
multilingual
summarizatio
n
(
ML
SUM
)
,
which
is
a
large
-
scale
Turkish
news
dataset,
and
success
was
exa
mined.
Then,
success
was
evaluated
by
first
applying
BERT
ext
ractive
summarization
and
then
abstractive
summarization
on
1010
articles
collected
on
the
Dergipark
site.
Rouge
measures
were
used
for
perfor
mance
evaluatio
n.
This
study
is
one
of
the
first
examples
in
the
Turkish
la
nguage
and it
is cons
idered to
provid
e a basis
for futu
re studi
es with
good res
ults.
K
e
y
w
o
r
d
s
:
A
bs
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on
D
a
ta
s
e
t
D
e
e
p l
e
a
r
ni
ng
P
r
e
-
tr
a
in
e
d
T
ur
ki
s
h t
e
xt
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
N
e
da
A
li
pour
D
e
pa
r
tm
e
nt
of
M
a
na
ge
m
e
nt
I
nf
or
m
a
ti
on S
ys
te
m
s
, F
a
c
ul
ty
of
E
c
onomi
c
s
a
nd A
dm
in
is
tr
a
ti
ve
S
c
ie
nc
e
A
ta
tü
r
k U
ni
ve
r
s
it
y
E
r
z
ur
um
25240
, T
ür
ki
ye
E
m
a
il
:
ne
da
.a
li
pour
14@
ogr
.a
ta
uni
.e
du.t
r
1.
I
N
T
R
O
D
U
C
T
I
O
N
W
it
h
th
e
a
dve
nt
of
th
e
in
te
r
ne
t
in
th
e
di
gi
ta
l
a
ge
,
th
e
r
e
ha
s
be
e
n
a
m
a
s
s
iv
e
in
c
r
e
a
s
e
in
a
c
c
e
s
s
to
te
xt
ua
l
in
f
or
m
a
ti
on.
A
ut
om
a
ti
c
te
xt
s
um
m
a
r
iz
a
ti
on,
w
hi
c
h
is
o
ne
of
th
e
di
f
f
e
r
e
nt
na
tu
r
a
l
la
ngua
ge
pr
oc
e
s
s
in
g
(
N
L
P
)
ta
s
ks
,
he
lp
s
to
obt
a
in
m
or
e
c
om
pa
c
t
a
nd
e
f
f
ic
ie
nt
v
e
r
s
io
ns
of
te
xt
c
ont
e
nt
in
a
s
hor
te
r
ti
m
e
by
obt
a
in
in
g t
he
m
os
t
im
por
ta
nt
i
nf
or
m
a
ti
on
[
1
]
, [
2
]
. T
hus
, i
t
w
a
s
t
r
ie
d t
o ove
r
c
om
e
t
he
d
if
f
ic
ul
ti
e
s
t
ha
t
e
m
e
r
ge
d
w
it
h
th
e
in
c
r
e
a
s
e
in
da
ta
.
W
it
h
th
e
in
c
r
e
a
s
e
in
d
a
ta
a
nd
due
to
r
e
pe
ti
ti
ve
a
nd
ir
r
e
le
va
nt
c
ont
e
nt
,
it
is
n
e
c
e
s
s
a
r
y
to
s
pe
nd
m
or
e
ti
m
e
a
nd
e
f
f
or
t
to
obt
a
in
im
por
ta
nt
in
f
or
m
a
ti
o
n
by
hum
a
ns
.
F
or
th
is
r
e
a
s
on,
a
ut
om
a
ti
c
te
xt
s
um
m
a
r
iz
a
ti
on
ha
s
be
e
n
one
of
th
e
is
s
u
e
s
th
a
t
s
houl
d
be
s
tu
di
e
d
a
nd
una
voi
da
bl
e
la
te
ly
.
F
or
a
ut
om
a
ti
c
te
xt
s
um
m
a
r
iz
a
ti
on,
te
xt
r
e
tr
ie
va
l
s
ys
te
m
s
a
r
e
us
e
d
to
di
s
pl
a
y
a
s
u
m
m
a
r
iz
e
d
ve
r
s
io
n
of
s
e
a
r
c
h
r
e
s
ul
ts
in
s
e
a
r
c
h
e
ngi
ne
s
[
3]
.
A
c
c
or
di
ng
to
th
e
M
or
e
no
[
4]
li
s
t,
te
xt
s
um
m
a
r
iz
a
ti
on
c
a
n
be
vi
e
w
e
d
f
r
om
di
f
f
e
r
e
nt
a
ngl
e
s
,
in
c
lu
di
ng
s
in
gl
e
-
doc
um
e
nt
[
5]
a
nd
m
ul
ti
-
doc
um
e
nt
[
6]
,
[
7]
in
te
r
m
s
of
num
be
r
of
in
put
doc
um
e
nt
s
,
m
onol
in
gua
l
a
nd
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
2
,
A
pr
il
20
2
5
:
1587
-
1596
1588
m
ul
ti
li
ngua
l
[
8]
in
te
r
m
s
of
num
be
r
o
f
in
put
la
ngua
ge
s
,
a
nd
e
xt
r
a
c
ti
ve
,
a
bs
tr
a
c
ti
ve
a
nd
hybr
id
s
um
m
a
r
iz
a
ti
on
in
te
r
m
s
of
out
put
ge
ne
r
a
ti
on
a
pp
r
oa
c
h:
e
xt
r
a
c
ti
ve
s
um
m
a
r
ie
s
de
f
in
e
th
e
c
a
ndi
da
te
s
e
nt
e
nc
e
s
a
c
c
or
di
ng
to
f
e
a
tu
r
e
s
s
uc
h a
s
t
he
l
e
ngt
h of
t
he
s
e
nt
e
nc
e
, t
he
pos
it
io
n of
t
he
s
e
nt
e
nc
e
s
r
e
la
ti
ve
t
o e
a
c
h ot
he
r
,
a
nd
th
e
r
a
ti
o of
nouns
,
by
c
r
e
a
ti
ng
a
s
e
nt
e
nc
e
s
c
or
in
g
m
e
c
ha
ni
s
m
f
or
th
e
m
os
t
im
por
ta
nt
s
e
nt
e
nc
e
s
f
r
om
th
e
s
our
c
e
a
nd
c
om
bi
ne
s
th
e
m
to
f
or
m
th
e
s
um
m
a
r
y.
C
a
ndi
da
te
s
e
nt
e
nc
e
s
a
r
e
r
a
nke
d
a
c
c
or
di
ng
to
th
e
s
pe
c
if
ie
d
f
e
a
tu
r
e
s
a
nd
s
c
or
in
g,
a
nd
th
e
c
a
ndi
da
te
s
e
nt
e
nc
e
s
a
t
th
e
to
p
of
th
e
r
e
qui
r
e
m
e
nt
a
r
e
s
e
l
e
c
te
d
[
9]
.
I
n
a
bs
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on,
ne
w
e
xpr
e
s
s
io
n
s
a
r
e
pr
oduc
e
d
w
it
h
s
e
nt
e
nc
e
s
th
a
t
a
r
e
not
f
ound
in
th
e
or
ig
in
a
l
te
xt
a
nd
a
r
e
tr
ie
d
to
obt
a
in
a
s
um
m
a
r
y
by
li
ngui
s
ti
c
m
e
th
ods
th
a
t
a
r
e
us
e
d
f
or
unde
r
s
ta
ndi
ng
a
nd
e
xa
m
in
in
g
th
e
te
xt
[
10]
,
[
11
]
.
A
bs
tr
a
c
ti
ve
te
xt
s
um
m
a
r
ie
s
a
r
e
m
or
e
a
tt
r
a
c
ti
ve
,
by
us
in
g
of
c
om
pl
e
x
na
tu
r
a
l
la
ngua
ge
c
om
pr
e
he
ns
io
n
a
nd
r
e
nde
r
in
g
c
a
pa
bi
li
ti
e
s
to
pr
oduc
e
hum
a
n
-
li
ke
s
um
m
a
r
ie
s
.
T
he
r
e
f
or
e
,
in
r
e
c
e
nt
y
e
a
r
s
,
a
bs
tr
a
c
ti
ve
te
c
hni
que
s
in
di
f
f
e
r
e
nt
la
ngua
ge
s
ha
ve
a
tt
r
a
c
te
d
m
or
e
a
tt
e
nt
io
n
w
it
h
a
dva
nc
e
s
in
de
e
p
le
a
r
ni
ng.
S
in
c
e
te
xt
s
um
m
a
r
iz
a
ti
on
c
a
n
b
e
s
e
e
n
a
s
a
s
e
que
nc
e
-
to
-
s
e
que
nc
e
(
S
e
q2S
e
q
)
ta
s
k,
th
e
r
e
a
r
e
di
f
f
e
r
e
nt
a
ppr
oa
c
he
s
to
a
bs
tr
a
c
t
te
xt
s
um
m
a
r
iz
a
ti
on,
e
s
pe
c
ia
ll
y
f
or
th
e
E
ngl
is
h
la
ngua
ge
.
T
e
xt
s
um
m
a
r
iz
a
ti
on
is
a
S
e
q2S
e
q
ta
s
k.
E
nc
ode
r
-
de
c
ode
r
a
r
c
hi
te
c
tu
r
e
-
ba
s
e
d
S
e
q2S
e
q
m
ode
ls
ha
ve
ga
in
e
d
s
ig
ni
f
ic
a
nt
a
tt
e
nt
io
n
in
r
e
c
e
nt
ye
a
r
s
.
A
s
a
r
e
s
ul
t,
th
e
r
e
ha
s
be
e
n
a
s
hi
f
t
f
r
om
lo
ng
s
hor
t
-
te
r
m
m
e
m
or
y
(
L
S
T
M
)
-
ba
s
e
d
m
ode
ls
[
12]
to
tr
a
ns
f
or
m
e
r
-
ba
s
e
d m
ode
ls
i
n e
nc
ode
r
-
de
c
ode
r
ne
twor
ks
[
13]
.
I
n
r
e
c
e
nt
ye
a
r
s
,
s
tu
di
e
s
ha
ve
s
how
n
ve
r
y
good
pe
r
f
or
m
a
nc
e
us
in
g
pr
e
-
tr
a
in
in
g
S
e
q2S
e
q
m
ode
ls
on
ve
r
y
la
r
ge
da
ta
s
e
ts
to
im
pr
ove
t
e
xt
s
um
m
a
r
iz
a
ti
on
[
14]
–
[
17]
a
nd
a
c
hi
e
vi
ng
s
ta
te
-
of
-
th
e
-
a
r
t
r
e
s
ul
ts
in
ne
ur
a
l
a
bs
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on
[
18]
.
U
nf
or
tu
na
te
ly
,
m
os
t
of
th
e
r
e
s
e
a
r
c
h
ha
s
be
e
n
done
in
E
ngl
is
h
onl
y,
a
nd
m
ode
ls
th
a
t
ne
e
d
pr
e
-
tr
a
in
e
d
r
e
qui
r
e
la
r
ge
a
m
ount
s
of
da
ta
a
nd
c
om
put
a
ti
ona
l
pow
e
r
.
H
ow
e
ve
r
,
r
e
c
e
nt
ly
m
ul
ti
li
ngua
l
ve
r
s
io
ns
of
b
id
ir
e
c
ti
ona
l
e
nc
ode
r
r
e
pr
e
s
e
nt
a
ti
ons
f
r
om
tr
a
ns
f
or
m
e
r
s
(
B
E
R
T
)
[
19]
th
e
w
id
e
ly
us
e
d
two
pr
e
-
tr
a
in
e
d
m
ul
ti
l
in
gua
l
S
e
q2S
e
q
m
ode
ls
m
ul
ti
li
ngu
a
l
te
xt
-
to
-
te
xt
t
r
a
ns
f
e
r
tr
a
ns
f
or
m
e
r
(
m
T
5
)
[
20]
a
nd
m
ul
ti
li
ngua
l
bi
di
r
e
c
ti
ona
l
a
nd
a
ut
o
-
r
e
gr
e
s
s
iv
e
tr
a
n
s
f
or
m
e
r
s
(
m
B
A
R
T
)
[
19]
,
[
21]
ha
ve
le
d
to
s
tu
di
e
s
in
s
e
ve
r
a
l
r
e
s
e
a
r
c
h
a
r
e
a
s
f
or
lo
w
-
r
e
s
our
c
e
la
ngua
ge
s
.
T
he
m
T
5
[
20]
m
ode
l,
w
hi
c
h
c
ove
r
s
101
di
f
f
e
r
e
nt
la
ngua
ge
s
a
nd
is
tr
a
in
e
d
on
a
c
om
m
on
la
ngua
g
e
,
is
a
m
ul
ti
li
ngua
l
ve
r
s
io
n
of
th
e
te
xt
-
to
-
te
xt
tr
a
ns
f
e
r
c
onve
r
te
r
(
T
5)
m
ode
l.
T
he
m
T
5 m
ode
l
i
s
a
s
ui
ta
bl
e
opt
io
n f
or
m
os
t
la
ngua
ge
s
du
e
t
o i
ts
m
ul
ti
li
ngua
l
f
e
a
tu
r
e
.
I
t
s
e
e
m
s
th
a
t
f
or
T
ur
ki
s
h
la
ngua
g
e
e
xt
r
a
c
ti
ve
a
ut
om
a
ti
c
te
xt
s
u
m
m
a
r
iz
a
ti
on
s
tu
di
e
s
ha
ve
be
e
n
done
m
or
e
.
H
ow
e
ve
r
,
th
e
r
e
a
r
e
ve
r
y
f
e
w
s
tu
di
e
s
f
oc
us
e
d
on
a
b
s
tr
a
c
t
iv
e
te
xt
s
um
m
a
r
iz
a
ti
on
f
or
T
ur
ki
s
h
[
22]
,
[
23]
.
I
n
th
e
s
e
s
tu
di
e
s
,
pr
e
-
tr
a
in
e
d
S
e
q2S
e
q
m
ode
ls
w
e
r
e
us
e
d
le
s
s
f
r
e
que
nt
ly
.
P
r
e
vi
ous
s
tu
di
e
s
in
N
L
P
ha
ve
de
m
ons
tr
a
te
d
th
a
t
te
c
hni
que
s
de
ve
lo
p
e
d
f
or
la
ngua
g
e
s
li
k
e
E
ng
li
s
h
of
te
n
pe
r
f
or
m
poor
ly
on
m
or
phol
ogi
c
a
ll
y
r
ic
h
la
ngua
ge
s
,
s
uc
h
a
s
T
ur
ki
s
h.
T
hi
s
hi
ghl
ig
ht
s
th
e
ne
e
d
f
or
a
ddi
ti
ona
l
m
e
th
ods
th
a
t
a
c
c
ount
f
or
th
e
uni
que
m
or
phol
ogi
c
a
l
s
tr
uc
tu
r
e
s
i
n
th
e
s
e
l
a
ngua
ge
s
[
24]
. F
or
e
xa
m
pl
e
, T
ur
ki
s
h i
s
a
n a
ggl
ut
in
a
ti
ve
l
a
ngua
ge
i
n
w
hi
c
h
r
oot
w
or
ds
c
a
n
a
c
qui
r
e
num
e
r
ous
de
r
iv
a
ti
ve
s
a
nd
in
f
le
c
ti
ons
.
T
hi
s
c
h
a
r
a
c
te
r
is
ti
c
r
e
s
ul
t
s
in
a
w
id
e
va
r
ie
ty
of
uni
que
w
or
d s
ur
f
a
c
e
f
or
m
s
, l
e
a
di
ng t
o c
ha
ll
e
nge
s
w
it
h da
ta
s
pa
r
s
it
y
[
25]
.
S
e
c
ti
on
2
of
th
is
a
r
ti
c
le
s
um
m
a
r
iz
e
s
r
e
la
te
d
w
or
k
a
nd
th
e
ir
a
c
hi
e
ve
m
e
nt
s
.
S
e
c
ti
ons
3
a
nd
4
pr
ovi
de
a
n
ove
r
vi
e
w
of
th
e
da
ta
s
e
ts
a
nd
th
e
m
T
5
m
e
th
od.
F
in
a
ll
y,
s
e
c
ti
ons
5
a
nd
6
c
onc
lu
de
th
e
a
r
ti
c
le
by
pr
e
s
e
nt
in
g
th
e
c
onc
lu
s
io
ns
of
t
he
s
tu
dy a
nd s
ugge
s
ti
ons
f
or
f
ut
ur
e
s
tu
di
e
s
.
2.
R
E
L
A
T
E
D
WORK
2.1.
P
r
e
-
t
r
ai
n
e
d
s
e
q
u
e
n
c
e
-
to
-
s
e
q
u
e
n
c
e
m
od
e
ls
I
n
r
e
c
e
nt
ye
a
r
s
,
s
ta
te
-
of
-
th
e
-
a
r
t
r
e
s
ul
ts
of
tr
a
ns
f
e
r
le
a
r
ni
ng
in
N
L
P
,
w
hi
c
h
ha
s
be
e
n
ve
r
y
e
f
f
e
c
ti
ve
,
ha
ve
e
m
e
r
ge
d
in
di
f
f
e
r
e
nt
ta
s
k
s
.
I
t
ha
s
be
e
n
de
te
r
m
in
e
d
th
a
t
th
e
c
onc
e
pt
of
pr
e
-
tr
a
in
in
g
of
a
la
ngua
ge
m
ode
l
th
a
t
c
a
n
le
a
r
n
ta
s
k
-
a
gnos
ti
c
kno
w
le
dge
in
na
tu
r
a
l
l
a
ngua
ge
c
o
m
pr
e
he
ns
io
n
a
nd
th
e
n
tr
a
n
s
f
e
r
it
to
s
ubs
e
que
nt
ta
s
ks
is
s
uc
c
e
s
s
f
ul
[
19]
,
[
26]
,
[
27]
.
H
ow
e
ve
r
,
ne
w
r
e
s
e
a
r
c
h
is
tu
r
ni
ng
to
pr
e
-
tr
a
in
e
d
S
e
q2S
e
q
m
ode
ls
be
c
a
us
e
pr
e
-
tr
a
in
e
d
e
nc
ode
r
m
ode
ls
do
not
w
or
k
w
e
ll
f
or
ta
s
ks
th
a
t
r
e
qui
r
e
na
tu
r
a
l
la
ngua
ge
g
e
ne
r
a
ti
on
a
nd
na
tu
r
a
l
la
ngua
ge
unde
r
s
ta
ndi
ng,
s
u
c
h
a
s
te
xt
s
um
m
a
r
iz
a
ti
on
a
nd
m
a
c
hi
ne
tr
a
ns
la
ti
on.
S
ong
e
t
al
.
[
15]
pr
opos
e
d
m
a
s
ke
d
s
e
que
nc
e
to
s
e
que
n
c
e
pr
e
-
tr
a
in
in
g
(
M
A
S
S
)
w
it
h
h
e
lp
f
r
om
B
E
R
T
in
r
e
c
ons
tr
uc
ti
ng
th
e
r
e
s
t
of
th
e
s
e
nt
e
nc
e
to
c
r
e
a
te
a
n
e
nc
od
e
r
-
de
c
ode
r
ba
s
e
d
la
ngu
a
ge
.
A
s
e
nt
e
nc
e
c
ont
a
in
in
g
a
r
a
ndoml
y
m
a
s
ke
d
pa
r
t
in
th
e
e
nc
ode
r
pa
r
t
is
us
e
d
a
s
in
put
a
nd
th
e
de
c
ode
r
p
a
r
t
tr
ie
s
to
gue
s
s
th
is
m
a
s
k
e
d
pa
r
t.
T
hus
,
th
e
M
A
S
S
m
ode
l
c
a
n
tr
a
in
th
e
e
nc
ode
r
a
nd
de
c
ode
r
to
ge
th
e
r
.
D
ong
e
t
al
.
[
14]
in
tr
oduc
e
d
a
ne
w
pr
e
-
tr
a
in
e
d
uni
f
ie
d
l
a
ngua
ge
m
od
e
l
(
U
ni
L
M
)
th
a
t
in
c
or
por
a
te
s
bi
di
r
e
c
ti
ona
l,
uni
di
r
e
c
ti
ona
l,
a
nd
S
e
q2
S
e
q
pr
e
di
c
ti
ve
la
ngua
g
e
m
ode
li
ng
ta
s
ks
.
T
hi
s
m
ode
l
c
a
n
be
f
in
e
-
tu
ne
d
f
or
ta
s
ks
in
vol
vi
ng
bot
h
unde
r
s
ta
ndi
ng
a
nd
ge
ne
r
a
ti
on
of
na
tu
r
a
l
la
ngua
ge
.
U
ni
L
M
e
m
e
r
ge
d
u
s
in
g
a
s
h
a
r
e
d
tr
a
ns
f
or
m
e
r
ne
twor
k a
nd
us
in
g
c
e
r
ta
in
s
e
lf
-
a
tt
e
nt
io
n m
a
s
ks
t
o c
ont
r
ol
c
ont
e
xt
in
w
hi
c
h
th
e
pr
e
di
c
ti
on
c
ondi
t
io
ns
a
r
e
.
L
e
w
is
e
t
al
.
[
17]
tr
a
in
e
d
w
it
h
bi
di
r
e
c
ti
ona
l
a
nd
a
ut
o
-
r
e
gr
e
s
s
iv
e
tr
a
ns
f
or
m
e
r
s
(
B
A
R
T
)
, one
of
t
he
a
ut
oe
nc
ode
r
S
eq
2S
eq
m
ode
ls
to
ge
ne
r
a
te
ne
w
t
e
xt
by f
ir
s
t
di
s
to
r
ti
ng t
he
t
e
xt
a
nd
th
e
n
le
a
r
ni
ng
a
m
ode
l.
F
or
th
is
pur
pos
e
,
th
e
y
us
e
d
a
s
ta
nda
r
d
tr
a
ns
f
or
m
e
r
-
ba
s
e
d
ne
ur
a
l
m
a
c
hi
ne
tr
a
ns
la
ti
on
a
r
c
hi
te
c
tu
r
e
.
F
in
e
-
tu
ni
ng
B
A
R
T
ha
s
s
how
n
to
b
e
e
f
f
e
c
ti
ve
f
or
te
xt
c
r
e
a
ti
on
a
nd
c
om
pr
e
he
n
s
io
n
ta
s
ks
.
R
a
f
f
e
l
e
t
al
.
[
28]
pr
ovi
de
d
a
n
ove
r
vi
e
w
of
tr
a
ns
f
e
r
le
a
r
ni
ng
te
c
hni
que
s
f
or
N
L
P
,
a
ls
o
th
e
y
c
om
pa
r
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
A
bs
tr
ac
ti
v
e
s
um
m
a
r
iz
at
io
n us
in
g m
ul
ti
li
ngual t
e
x
t
-
to
-
te
x
t
tr
ans
fe
r
t
r
ans
fo
r
m
e
r
f
o
r
t
he
…
(
N
e
da A
li
pour
)
1589
pr
e
-
tr
a
in
in
g
goa
ls
,
tr
a
ns
f
e
r
a
ppr
oa
c
h
e
s
,
a
r
c
hi
te
c
tu
r
e
s
,
unl
a
be
le
d
da
ta
s
e
ts
a
nd
ot
he
r
f
a
c
to
r
s
f
or
la
ngua
ge
unde
r
s
ta
ndi
ng.
X
u
e
e
t
a
l
.
[
20]
i
nt
r
o
du
c
e
d
m
T
5,
a
p
r
e
-
tr
a
in
e
d
m
ul
ti
l
in
g
ua
l
va
r
ia
nt
of
T
5
a
v
a
i
la
bl
e
i
n
101
la
n
gu
a
g
e
s
,
o
n
a
n
e
w
c
o
m
m
o
n
s
c
a
n
-
b
a
s
e
d
d
a
t
a
s
e
t.
I
t
d
e
m
on
s
tr
a
t
e
d
s
t
a
t
e
-
of
-
t
h
e
-
a
r
t
pe
r
f
or
m
a
n
c
e
in
m
ul
t
il
i
ngu
a
l
be
n
c
h
m
a
r
k
s
by
d
e
t
a
il
in
g
t
he
m
T
5'
s
m
od
if
i
e
d
tr
a
i
ni
n
g
a
n
d
de
s
i
gn.
L
i
u
e
t
a
l
.
[
21
]
f
oun
d
th
a
t
m
u
lt
i
li
n
gu
a
l
de
n
oi
s
i
ng
pr
e
-
tr
a
in
in
g
f
or
a
w
i
de
r
a
ng
e
of
m
a
c
h
in
e
t
r
a
ns
la
ti
o
n yi
e
ld
s
hu
g
e
p
e
r
f
or
m
a
n
c
e
g
a
i
ns
.
D
e
vl
i
n
e
t
a
l
.
[
19]
of
f
e
r
e
d m
B
A
R
T
i
ns
pi
r
e
d by B
A
R
T
t
o pr
e
-
tr
a
in
t
he
S
e
q2S
e
q
m
ode
l.
2.2.
A
b
s
t
r
ac
t
iv
e
t
e
xt
s
u
m
m
a
r
iz
at
io
n
W
it
h
th
e
de
ve
lo
pm
e
nt
of
de
e
p
le
a
r
ni
ng,
e
nc
ode
r
de
c
od
e
r
S
e
q2
S
e
q
ne
twor
ks
ha
ve
s
ta
r
te
d
to
ga
in
m
or
e
im
por
ta
nc
e
f
or
a
bs
tr
a
c
ti
ve
te
xt
s
um
m
a
r
iz
a
ti
on.
R
us
h
e
t
al
.
[
29]
pr
opos
e
d
a
ne
ur
a
l
ne
twor
k
la
ngua
ge
m
ode
l
(
N
N
L
M
)
,
a
ne
ur
a
l
lo
c
a
l
a
tt
e
nt
io
n
-
ba
s
e
d
m
ode
l
th
a
t
c
a
n
be
e
a
s
il
y
tr
a
in
e
d
a
nd
s
c
a
la
bl
e
to
tr
a
in
in
g
da
ta
f
or
a
bs
tr
a
c
ti
ve
s
e
nt
e
nc
e
s
um
m
a
r
iz
a
ti
on.
C
hopr
a
e
t
al
.
[
10]
pr
opos
e
d
a
c
onvolut
io
na
l
a
tt
e
nt
io
n
-
ba
s
e
d
e
n
c
ode
r
m
ode
l
a
s
a
s
im
pl
if
ie
d
ve
r
s
io
n
of
th
e
e
nc
od
e
r
-
de
c
ode
r
f
r
a
m
e
w
or
k
us
in
g
a
r
e
c
ur
r
e
nt
ne
ur
a
l
ne
twor
k
(
R
N
N
)
f
or
a
bs
tr
a
c
ti
ve
s
e
nt
e
nc
e
s
um
m
a
r
iz
a
ti
on.
I
t
w
a
s
us
e
d
two
-
la
ye
r
L
S
T
M
s
f
or
th
e
e
nc
ode
r
-
de
c
ode
r
c
ont
a
in
in
g
500
hi
dde
n
uni
ts
in
e
a
c
h
la
ye
r
.
N
a
ll
a
pa
ti
e
t
al
.
[
11]
pr
opos
e
d
a
n
e
w
da
ta
s
e
t
of
m
ul
ti
-
s
e
nt
e
nc
e
s
um
m
a
r
ie
s
a
nd
s
e
ve
r
a
l
ne
w
m
od
e
ls
f
or
a
bs
tr
a
c
ti
ve
te
xt
s
um
m
a
r
iz
a
ti
on
u
s
in
g
bi
di
r
e
c
ti
ona
l
L
S
T
M
-
ba
s
e
d
e
nc
ode
r
-
de
c
ode
r
,
s
uc
h
a
s
f
e
a
tu
r
e
r
ic
h
e
nc
ode
r
,
m
ode
li
ng
ke
yw
or
ds
,
a
nd
a
hi
e
r
a
r
c
hi
c
a
l
e
nc
ode
r
-
de
c
ode
r
th
a
t
is
c
a
pa
bl
e
of
c
a
pt
ur
in
g
th
e
doc
um
e
nt
s
tr
uc
tu
r
e
.
C
e
li
kyi
lm
a
z
e
t
al
.
[
30]
e
xt
e
nde
d
th
e
C
om
m
N
e
t
m
ode
l
of
[
31]
on
C
N
N
/Da
il
yM
a
il
a
nd
N
e
w
Y
or
k
T
im
e
s
da
ta
s
e
ts
f
or
a
bs
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on
w
it
h
de
e
p
c
om
m
uni
c
a
ti
on
a
ge
nt
s
in
a
n e
nc
ode
r
-
de
c
ode
r
a
r
c
hi
te
c
tu
r
e
.
P
a
ul
us
e
t
al
.
[
32]
in
tr
oduc
e
d
a
n
e
w
tr
a
in
in
g
m
e
th
od
f
or
a
bs
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on
th
a
t
c
om
bi
ne
s
s
ta
nda
r
d
s
upe
r
vi
s
e
d
w
or
d
pr
e
di
c
ti
on
a
nd
a
ne
ur
a
l
ne
twor
k
m
ode
l
w
it
h
r
e
in
f
or
c
e
m
e
nt
le
a
r
ni
ng
(
R
L
)
on
th
e
C
N
N
/Da
il
yM
a
il
da
ta
s
e
t.
N
a
r
a
ya
n
e
t
al
.
[
33]
pr
opos
e
d
e
xt
r
e
m
e
s
um
m
a
r
iz
a
ti
on
ba
s
e
d
on
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
on
a
l
a
r
ge
-
s
c
a
le
da
ta
s
e
t
by
c
ol
le
c
ti
ng
onl
in
e
a
r
ti
c
le
s
f
r
om
th
e
B
r
it
is
h
B
r
oa
dc
a
s
ti
ng
C
or
por
a
ti
on
(
B
B
C
)
f
or
s
in
gl
e
-
doc
um
e
nt
a
bs
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on
a
nd
f
or
c
r
e
a
ti
ng
a
one
-
s
e
nt
e
nc
e
ne
w
s
s
um
m
a
r
y.
L
iu
a
nd
L
a
pa
ta
[
34]
pr
e
s
e
nt
e
d
a
ge
n
e
r
a
l
f
r
a
m
e
w
or
k
on
th
e
C
N
N
/Da
il
yM
a
il
n
e
w
s
hi
ghl
ig
ht
s
da
ta
s
e
t
[
35]
a
nd
th
e
N
e
w
Y
or
k
T
im
e
s
A
nnot
a
te
d
C
or
pus
[
36]
f
or
bot
h
e
xt
r
a
c
ti
ve
a
nd
a
b
s
tr
a
c
ti
ve
m
ode
l
s
,
a
nd
on
X
s
um
[
33]
f
or
th
e
B
E
R
T
-
ba
s
e
d
c
od
e
r
.
in
th
is
m
od
e
l,
a
n
e
w
f
in
e
-
tu
ni
ng pr
ogr
a
m
is
pr
opos
e
d f
or
a
bs
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on, whic
h a
dopt
s
di
f
f
e
r
e
nt
opt
im
iz
e
r
s
f
or
e
nc
ode
r
a
nd
de
c
ode
r
.
D
e
vl
in
e
t
al
.
[
19]
in
tr
oduc
e
d
B
A
R
T
,
a
pr
e
-
tr
a
in
in
g
a
ut
o
-
e
nc
ode
r
a
ppr
oa
c
h.
A
c
c
or
di
ng
to
th
e
a
ut
hor
s
,
B
A
R
T
w
or
ks
w
e
ll
f
or
te
xt
ge
ne
r
a
ti
on
a
nd
te
xt
c
om
pr
e
he
ns
io
n
ta
s
ks
w
he
n
f
in
e
-
tu
ne
d.
Z
ha
ng
e
t
al
.
[
18]
pr
opos
e
d
pr
e
-
tr
a
in
in
g
w
it
h
e
xt
r
a
c
te
d
g
a
p
-
s
e
nt
e
nc
e
s
f
or
a
bs
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on
(
P
E
G
A
S
U
S
)
,
w
hi
c
h
pr
e
-
t
r
a
in
e
d
th
e
la
r
ge
tr
a
ns
f
or
m
e
r
-
ba
s
e
d
e
nc
od
e
r
-
de
c
ode
r
f
or
a
bs
tr
a
c
ti
ve
te
xt
s
um
m
a
r
iz
a
ti
on.
P
E
G
A
S
U
S
s
e
le
c
ts
a
nd
m
a
s
k
s
im
por
ta
nt
s
e
nt
e
n
c
e
s
in
th
e
do
c
um
e
nt
a
nd
c
r
e
a
t
e
s
ga
p
s
e
nt
e
n
c
e
s
a
s
a
pr
e
-
tr
a
in
in
g
ta
r
ge
t.
T
h
e
y
e
va
lu
a
te
d
th
e
be
s
t
P
E
G
A
S
U
S
m
ode
ls
f
or
12
dow
ns
tr
e
a
m
s
um
m
a
r
ie
s
c
ov
e
r
in
g
s
c
ie
nc
e
,
n
e
w
s
,
s
to
r
ie
s
,
in
s
tr
uc
ti
ons
,
pa
te
nt
s
,
e
m
a
il
s
,
a
nd
bi
ll
s
.
Q
i
e
t
al
.
[
37]
in
tr
oduc
e
d
a
ne
w
s
e
lf
-
s
upe
r
vi
s
e
d
obj
e
c
ti
ve
,
f
ut
ur
e
n
-
gr
a
m
pr
e
di
c
ti
on,
w
hi
c
h
w
a
s
te
s
te
d
on
th
e
C
N
N
/Da
il
yM
a
il
,
G
ig
a
w
or
d,
a
nd
S
Q
uA
D
1.1
be
nc
hm
a
r
ks
f
or
ta
s
ks
li
ke
que
s
ti
on
ge
ne
r
a
ti
on
a
nd
a
b
s
tr
a
c
t
iv
e
s
um
m
a
r
iz
a
ti
on.
T
he
y
a
ls
o
de
v
e
lo
pe
d
a
S
e
q2
S
e
q
pr
e
-
tr
a
in
e
d
m
ode
l
c
a
ll
e
d
P
r
ophe
tNe
t,
f
e
a
tu
r
in
g
a
n
n
-
s
tr
e
a
m
s
e
lf
-
a
tt
e
nt
io
n
m
e
c
h
a
ni
s
m
.
I
n
c
ont
r
a
s
t
to
c
onve
nt
io
na
l
S
e
q2S
e
q
m
ode
ls
,
P
r
ophe
tNe
t
is
opt
im
iz
e
d
f
or
n
-
s
te
p
f
or
w
a
r
d
pr
e
di
c
ti
on,
pr
e
di
c
ti
ng
th
e
ne
xt
n
to
ke
ns
ba
s
e
d
on
pr
e
vi
ous
c
ont
e
xt
to
ke
n
s
a
t
e
a
c
h
ti
m
e
s
te
p.
P
r
ophe
tNe
t
w
a
s
pr
e
-
tr
a
in
e
d
on
bot
h
a
ba
s
e
-
s
c
a
le
da
ta
s
e
t
(
16
G
B
)
a
nd a
l
a
r
ge
-
s
c
a
le
da
ta
s
e
t
(
160
G
B
)
.
2.3.
T
u
r
k
is
h
t
e
xt
s
u
m
m
a
r
iz
at
io
n
In
s
tu
dy
by
A
lt
a
n
[
38]
,
th
e
s
ys
te
m
w
a
s
de
ve
lo
p
e
d
by
s
in
gl
e
T
ur
ki
s
h
doc
um
e
nt
a
s
in
put
a
nd
s
c
or
in
g
w
a
s
c
a
r
r
ie
d
out
us
in
g
f
e
a
tu
r
e
s
s
uc
h
a
s
s
e
nt
e
n
c
e
lo
c
a
ti
on
in
f
o
r
m
a
ti
on
a
nd
te
r
m
f
r
e
que
nc
y
in
f
or
m
a
ti
on,
a
n
d
s
um
m
a
r
ie
s
w
e
r
e
obt
a
in
e
d
us
in
g
a
num
be
r
of
s
ta
ti
s
ti
c
a
l
m
e
th
o
ds
.
K
ut
lu
e
t
al
.
[
39]
pr
opos
e
d
a
ge
ne
r
a
l
te
xt
s
um
m
a
r
iz
a
ti
on
m
e
th
od
ba
s
e
d
on
s
e
nt
e
nc
e
or
de
r
in
g.
T
he
s
ys
te
m
c
a
lc
ul
a
t
e
d
s
e
nt
e
nc
e
s
c
or
e
s
us
in
g
s
ur
f
a
c
e
-
le
ve
l
f
e
a
tu
r
e
s
a
nd
pr
oduc
e
d
s
um
m
a
r
ie
s
by
s
e
le
c
ti
ng
th
e
hi
ghe
s
t
-
s
c
or
in
g
s
e
nt
e
n
c
e
s
f
r
om
th
e
or
ig
in
a
l
doc
um
e
nt
s
.
F
e
a
tu
r
e
s
s
uc
h
a
s
s
e
nt
e
n
c
e
pos
it
io
n,
ti
tl
e
s
im
il
a
r
it
y,
ke
y
phr
a
s
e
c
e
nt
r
a
li
ty
a
nd
a
nd
te
r
m
f
r
e
que
nc
y
w
e
r
e
a
ppl
ie
d.
T
he
s
tu
dy
e
m
pha
s
iz
e
d
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
c
e
nt
r
a
li
ty
a
s
a
f
e
a
tu
r
e
a
nd
w
a
s
one
of
th
e
f
ir
s
t
to
s
how
c
a
s
e
th
e
u
s
e
of
ke
y
phr
a
s
e
s
in
s
um
m
a
r
iz
in
g
T
ur
ki
s
h
te
xt
s
[
39]
,
[
40]
.
T
he
a
ut
hor
s
a
r
gue
d
th
a
t
th
e
c
r
os
s
m
e
th
od
de
ve
lo
pe
d
in
th
e
s
tu
dy
out
pe
r
f
or
m
e
d
ot
he
r
la
te
nt
s
e
m
a
nt
ic
a
na
ly
s
is
(
L
S
A
)
m
e
th
ods
.
O
z
s
oy
e
t
al
.
[
41]
in
tr
oduc
e
d
two
ne
w
L
S
A
-
ba
s
e
d
ha
s
hi
ng
a
lg
or
it
hm
s
a
nd
pr
e
s
e
nt
e
d
a
ge
ne
r
a
l
e
xt
r
a
c
ti
ve
te
xt
s
um
m
a
r
iz
a
ti
on
s
ys
te
m
f
or
T
ur
ki
s
h,
ba
s
e
d
on
L
S
A
.
P
e
m
be
[
42]
pr
opos
e
d
a
r
ul
e
-
ba
s
e
d
a
ppr
oa
c
h
f
or
a
ut
om
a
ti
c
doc
um
e
nt
s
um
m
a
r
iz
a
ti
on
ba
s
e
d
on
in
f
or
m
a
ti
on
r
e
que
s
ts
a
nd
te
xt
s
tr
u
c
tu
r
e
f
or
s
e
a
r
c
h
e
ngi
ne
s
.
A
f
te
r
s
c
or
in
g
th
e
s
e
nt
e
nc
e
s
u
s
in
g
th
e
po
s
it
io
n,
ti
tl
e
(
th
e
f
r
e
que
nc
y
of
oc
c
ur
r
e
nc
e
of
th
e
te
r
m
s
in
th
e
ti
tl
e
in
th
e
s
e
nt
e
n
c
e
)
,
que
r
y
s
e
nt
e
nc
e
a
nd
t
e
r
m
f
r
e
que
nc
y
m
e
th
ods
(
th
e
va
lu
e
obt
a
in
e
d
f
r
om
th
e
f
r
e
que
nc
y
of
oc
c
ur
r
e
nc
e
of
th
e
te
r
m
s
in
th
e
s
e
nt
e
n
c
e
in
th
e
w
hol
e
do
c
um
e
nt
)
,
s
c
or
e
s
w
e
r
e
gi
ve
n
a
c
c
o
r
di
ng
to
th
e
im
por
ta
nc
e
of
th
e
s
e
nt
e
nc
e
s
a
nd
s
e
nt
e
nc
e
s
e
l
e
c
ti
on
w
a
s
c
a
r
r
ie
d
out
.
G
ür
a
n
[
43]
pr
opos
e
d
a
ne
w
w
e
ig
ht
v
a
lu
e
f
or
e
xt
r
a
c
ti
ve
te
xt
s
um
m
a
r
iz
a
ti
on
th
a
t
c
a
n
be
us
e
d
in
te
xt
s
um
m
a
r
iz
a
ti
on
m
e
th
ods
ba
s
e
d
on
L
S
A
.
I
n
th
is
s
tu
dy,
a
hyb
r
id
s
ys
te
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
2
,
A
pr
il
20
2
5
:
1587
-
1596
1590
w
a
s
pr
opos
e
d
w
it
h
two
di
f
f
e
r
e
nt
a
ppr
oa
c
he
s
th
a
t
c
om
bi
ne
s
e
m
a
nt
ic
a
nd
s
tr
uc
tu
r
a
l
f
e
a
tu
r
e
s
f
or
im
por
ta
nt
s
e
nt
e
nc
e
e
xt
r
a
c
ti
on.
A
b
s
tr
a
c
ti
ve
te
xt
s
um
m
a
r
iz
a
ti
on
s
tu
di
e
s
us
i
ng
S
e
q2S
e
q
m
ode
ls
a
r
e
v
e
r
y
f
e
w
a
nd
li
m
it
e
d
f
or
T
ur
ki
s
h
te
xt
s
.
S
c
ia
lo
m
e
t
al
.
[
22]
pr
e
s
e
nt
e
d
m
ul
ti
li
ngua
l
s
um
m
a
r
iz
a
ti
on
(
M
L
S
U
M
)
,
th
e
f
ir
s
t
la
r
ge
-
s
c
a
l
e
M
L
S
U
M
da
ta
s
e
t,
in
f
iv
e
di
f
f
e
r
e
nt
la
ngua
ge
s
(
T
ur
ki
s
h,
R
us
s
ia
n,
S
pa
ni
s
h,
G
e
r
m
a
n,
a
nd
F
r
e
nc
h)
,
in
c
lu
di
ng
ove
r
1.5
m
il
li
on
a
r
ti
c
le
/s
um
m
a
r
y
pa
ir
s
f
r
om
onl
in
e
ne
w
s
pa
pe
r
s
,
to
e
va
lu
a
te
S
e
q2S
e
q
m
ode
ls
.
T
hi
s
s
tu
dy
r
e
por
ts
c
r
os
s
-
la
ngua
ge
c
om
pa
r
a
ti
ve
a
na
ly
s
is
ba
s
e
d
on
s
ta
te
-
of
-
th
e
-
a
r
t
s
ys
te
m
s
.
I
n
th
e
s
tu
dy
of
[
44]
,
a
n
e
nc
ode
r
-
de
c
ode
r
m
ode
l
w
a
s
de
ve
lo
pe
d
f
or
th
e
pr
e
di
c
ti
on
of
a
bs
tr
a
c
ti
ve
T
ur
ki
s
h
ne
w
s
he
a
dl
in
e
s
a
nd
th
e
s
ys
te
m
w
a
s
tr
a
in
e
d
w
it
h
R
N
N
.
F
a
s
tT
e
xt
m
ode
l
w
a
s
us
e
d
f
or
w
or
d
p
la
c
e
m
e
nt
in
ne
w
s
te
xt
s
.
B
a
yka
r
a
a
nd
G
üngör
[
23]
e
va
lu
a
te
d
s
e
ve
r
a
l
m
or
phol
ogi
c
a
l
to
k
e
ni
z
a
ti
on
m
e
th
ods
us
in
g
th
e
poi
nt
e
r
-
ge
ne
r
a
to
r
m
ode
l,
pr
e
s
e
nt
in
g
two
la
r
ge
-
s
c
a
l
e
da
ta
s
e
ts
(
H
U
-
N
e
w
s
a
nd
TR
-
N
e
w
s
)
to
ge
ne
r
a
te
a
b
s
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on
in
T
ur
ki
s
h
a
nd
H
ung
a
r
ia
n.
T
he
y
a
ls
o
c
om
p
a
r
e
d
th
e
r
e
s
ul
t
s
obt
a
in
e
d
f
r
om
th
e
T
R
-
N
e
w
s
d
a
ta
s
e
t
w
it
h
B
E
R
T
-
ba
s
e
d m
od
e
ls
.
3.
D
A
T
A
S
E
T
A
N
D
R
E
S
E
A
R
C
H
M
E
T
H
O
D
O
L
O
G
Y
I
n
th
e
te
xt
s
um
m
a
r
iz
a
ti
on
a
r
e
a
,
m
os
t
da
ta
s
e
ts
a
r
e
a
va
il
a
bl
e
in
E
ngl
is
h,
a
nd
da
ta
s
e
ts
in
ot
he
r
la
ngua
ge
s
s
uc
h
a
s
T
ur
ki
s
h
a
r
e
li
m
it
e
d.
I
n
th
is
s
tu
dy,
f
ir
s
tl
y,
th
e
M
L
S
U
M
[
21]
ne
w
s
da
ta
s
e
t
a
nd
th
e
n
th
e
a
r
ti
c
le
da
ta
s
e
t
c
r
e
a
te
d
by
th
e
a
ut
hor
w
e
r
e
us
e
d.
M
L
S
U
M
c
ov
e
r
s
5
la
ngu
a
ge
s
a
s
F
r
e
nc
h,
G
e
r
m
a
n,
S
pa
ni
s
h,
T
ur
ki
s
h
,
a
nd
R
us
s
ia
n a
nd
is
known a
s
t
e
xt
s
um
m
a
r
iz
a
ti
on
da
ta
s
e
t.
T
he
M
L
S
U
M
da
ta
s
e
t
w
a
s
c
r
e
a
te
d
f
r
om
th
e
popula
r
C
N
N
/Da
il
y
M
a
il
da
ta
s
e
t.
T
h
e
a
r
ti
c
le
da
ta
s
e
t
w
a
s
c
ol
le
c
te
d
f
r
om
th
e
D
e
r
gi
pa
r
k
a
nd
a
ll
s
ubj
e
c
ts
w
e
r
e
in
c
lu
de
d. T
he
r
e
a
r
e
1010 a
r
ti
c
le
s
i
n t
hi
s
da
ta
s
e
t.
T
hi
s
s
e
c
ti
on pr
ovi
de
s
a
n ove
r
vi
e
w
of
t
he
m
T
5
[
20]
a
r
c
hi
te
c
tu
r
e
t
ha
t
is
t
he
m
u
lt
il
in
gua
l
va
r
ia
nt
of
t
he
T
5
m
ode
l
[
28]
.
T
5
i
s
pr
e
-
tr
a
in
e
d
te
xt
-
to
-
te
xt
e
nc
ode
r
-
de
c
ode
r
t
r
a
ns
f
or
m
e
r
m
ode
l
w
hi
c
h
c
lo
s
e
ly
f
ol
lo
w
s
th
e
or
ig
in
a
ll
y
pr
opos
e
d
tr
a
ns
f
or
m
e
r
a
r
c
hi
te
c
tu
r
e
[
13]
a
nd
c
a
n
b
e
u
s
e
d
f
or
a
ll
te
xt
-
ba
s
e
d
N
L
P
pr
obl
e
m
s
[
20]
a
nd
c
ove
r
s
th
e
f
ol
lo
w
in
g
goa
ls
:
p
r
e
di
c
ti
ng
th
e
ne
xt
w
or
d
w
i
th
la
ngua
ge
m
ode
li
ng
,
r
e
de
f
in
in
g
th
e
or
ig
in
a
l
te
xt
w
it
h
d
e
-
s
huf
f
li
ng
, a
nd p
r
e
di
c
ti
ng ma
s
ke
d w
or
ds
w
it
h
c
or
r
upt
in
g s
pa
ns
[
5]
.
T
hi
s
a
ppr
oa
c
h
is
a
n
N
L
P
f
r
a
m
e
w
or
k
f
or
ge
ne
r
a
ti
ve
ta
s
ks
s
uc
h
a
s
te
xt
s
um
m
a
r
iz
a
ti
on,
que
s
ti
on
a
ns
w
e
r
in
g
a
nd
t
e
xt
c
la
s
s
if
ic
a
ti
on
w
he
r
e
th
e
ta
s
k
f
or
m
a
t
a
ll
ow
s
th
e
m
ode
l
to
ge
ne
r
a
te
te
xt
ba
s
e
d
on
s
om
e
in
put
[
20]
,
[
23]
.
A
s
a
r
e
s
ul
t,
th
e
s
a
m
e
hype
r
pa
r
a
m
e
te
r
s
a
nd
l
os
s
f
unc
ti
on
a
r
e
a
ppl
ie
d
a
c
r
os
s
e
a
c
h
ta
s
k
[
5]
.
F
ig
ur
e
1
il
lu
s
tr
a
te
s
th
e
T
5
m
ode
l
a
s
a
uni
f
ie
d
f
r
a
m
e
w
or
k
f
o
r
d
ow
ns
tr
e
a
m
N
L
P
ta
s
ks
.
E
a
c
h
dow
ns
tr
e
a
m
ta
s
k
in
te
xt
-
to
-
te
xt
f
or
m
a
t
is
r
e
pr
e
s
e
nt
e
d
by
a
di
f
f
e
r
e
nt
c
ol
or
:
tr
a
n
s
la
ti
on
(
gr
e
e
n)
,
li
ngui
s
ti
c
a
c
c
e
pt
a
bi
li
ty
(
r
e
d)
,
s
e
nt
e
nc
e
s
im
il
a
r
it
y
(
ye
ll
ow
)
,
a
nd
te
xt
s
um
m
a
r
iz
a
ti
on
(
bl
ue
)
[
28]
.
A
lt
hough
T
5
m
ode
l
w
a
s
tr
a
in
e
d
onl
y
f
or
E
ngl
is
h
la
ngua
ge
,
m
T
5
m
ode
l
w
a
s
tr
a
in
e
d
on
101
di
f
f
e
r
e
nt
l
a
ngua
ge
s
(
in
c
lu
di
ng
T
ur
ki
s
h)
a
nd
in
he
r
it
s
a
ll
c
a
pa
bi
li
ti
e
s
of
th
e
T
5
m
ode
l.
m
T
5
lo
oks
m
or
e
pow
e
r
f
ul
c
om
pa
r
e
d
to
ot
he
r
m
ode
ls
s
uc
h
a
s
B
E
R
T
,
c
r
os
s
-
li
ngua
l
la
ngua
ge
m
ode
l
w
it
h R
o
B
E
R
T
a
(
X
L
M
-
R
),
a
nd m
ul
ti
li
ngua
l
B
E
R
T
[
5]
.
F
ig
ur
e
1. mT
5 f
r
a
m
e
w
or
k
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
A
bs
tr
ac
ti
v
e
s
um
m
a
r
iz
at
io
n us
in
g m
ul
ti
li
ngual t
e
x
t
-
to
-
te
x
t
tr
ans
fe
r
t
r
ans
fo
r
m
e
r
f
o
r
t
he
…
(
N
e
da A
li
pour
)
1591
4.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
I
n
th
is
pa
pe
r
,
f
in
e
-
tu
ne
of
m
T
5
w
a
s
u
s
e
d
f
or
T
ur
ki
s
h
n
e
w
s
a
n
d
T
ur
ki
s
h
pa
pe
r
s
s
um
m
a
r
iz
a
ti
on.
W
e
us
e
d
A
da
m
opt
im
iz
e
r
,
8
a
nd
16
ba
tc
h
s
iz
e
a
nd
15
tr
a
in
in
g
e
p
oc
hs
a
s
f
in
e
-
tu
ne
pa
r
a
m
e
te
r
s
.
F
ir
s
t
of
a
ll
,
tr
a
in
in
g
w
a
s
c
a
r
r
ie
d
out
on
T
ur
ki
s
h
ne
w
s
w
it
h
8
a
nd
16
ba
tc
h
s
iz
e
s
a
nd
15
e
poc
hs
.
T
o
e
va
lu
a
te
th
e
m
ode
l,
th
e
r
e
s
ul
ts
w
e
r
e
e
va
lu
a
te
d
w
it
h
r
ouge
m
e
t
r
ic
s
.
R
ouge
m
e
tr
ic
s
m
os
t
c
om
m
onl
y
us
e
d
to
e
va
lu
a
te
te
x
t
s
um
m
a
r
iz
a
ti
on
a
nd
tr
a
n
s
la
ti
on
va
lu
e
s
.
I
n
th
i
s
s
tu
dy,
th
e
s
uc
c
e
s
s
of
th
e
m
ode
l
w
a
s
e
xa
m
in
e
d
w
it
h
obt
a
in
in
g
of
R
ouge
1, R
ouge
2
,
a
nd
R
ouge
L
.
R
ouge
-
N
i
s
a
m
e
th
od
th
a
t
s
c
o
r
e
s
th
e
s
e
ns
it
iv
it
y
v
a
lu
e
be
tw
e
e
n
th
e
r
e
f
e
r
e
nc
e
a
bs
tr
a
c
t
a
nd
th
e
c
a
ndi
da
te
a
bs
tr
a
c
t
a
c
c
or
di
ng
to
n
-
gr
a
m
ove
r
la
p.
I
t
t
r
ie
s
to
f
in
d
th
e
r
e
pe
ti
ti
on
r
a
te
of
th
e
pa
r
ts
di
vi
de
d
by
th
e
num
be
r
in
a
n N
-
gr
a
m
w
or
d
s
tr
in
g.
S
im
il
a
r
ly
,
R
ouge
-
L
va
lu
e
us
e
s
th
e
lo
nge
s
t
c
om
m
on
w
or
d
s
ubs
e
que
n
c
e
be
twe
e
n
two
di
f
f
e
r
e
nt
a
bs
tr
a
c
ts
[
45]
.
T
r
a
in
a
nd
va
li
da
ti
on
lo
s
s
f
or
T
ur
ki
s
h
ne
w
s
w
it
h
8
ba
tc
h
s
iz
e
w
e
r
e
s
how
n
in
F
ig
ur
e
2.
T
r
a
in
a
nd
va
li
da
ti
on
lo
s
s
f
or
T
ur
ki
s
h
ne
w
s
w
it
h
16
ba
tc
h
s
iz
e
w
e
r
e
s
how
n
in
F
ig
ur
e
3.
A
f
t
e
r
tr
a
i
n
in
g
t
he
m
o
d
e
l
w
i
th
b
a
t
c
h
s
iz
e
s
of
8
,
16
,
a
n
d
1
5
e
p
oc
h
s
,
t
h
e
m
o
s
t
c
o
m
m
on
l
y
u
s
e
d
R
ou
g
e
m
e
tr
ic
s
w
e
r
e
u
s
e
d
a
n
d
th
e
s
u
c
c
e
s
s
of
t
h
e
m
o
d
e
l
w
a
s
e
x
a
m
i
n
e
d
b
y
o
bt
a
in
in
g
R
o
ug
e
1
,
R
ou
g
e
2
,
a
nd
R
o
ug
e
L
.
R
o
u
g
e
-
1
,
R
o
u
ge
-
2
,
a
n
d
R
o
ug
e
-
L
v
a
l
u
e
s
w
e
r
e
s
ho
w
n
i
n
T
a
bl
e
1
.
T
h
e
s
u
c
c
e
s
s
a
c
hi
e
v
e
d
w
a
s
c
om
p
a
r
e
d
b
y
th
e
s
u
c
c
e
s
s
o
f
ot
he
r
s
tu
di
e
s
.
A
hu
ir
e
t
al
.
[
4
6]
w
or
k
e
d
o
n
a
b
s
tr
a
c
ti
v
e
s
u
m
m
a
r
i
z
a
t
io
n
w
it
h
m
T
5
i
n
S
p
a
ni
s
h
a
nd
C
a
t
a
l
a
n.
T
h
e
y
o
bt
a
in
e
d
t
h
e
f
ol
l
ow
in
g v
a
l
u
e
s
a
s
R
ou
g
e
-
1,
R
o
ug
e
-
2
,
a
n
d R
o
u
g
e
-
L
(
T
a
b
l
e
2)
.
P
a
n
t
a
nd
C
h
op
r
a
[
47
]
w
or
ke
d
on
s
um
m
a
ti
ve
s
um
m
a
r
iz
a
ti
on
w
it
h
m
T
5
in
S
pa
ni
s
h
a
nd
G
r
e
e
k
doc
um
e
nt
s
.
T
h
e
y
ju
s
t
e
va
lu
a
te
d
th
e
R
ouge
-
2
m
e
tr
ic
a
nd
got
13.1
f
or
S
pa
ni
s
h
a
nd
13.8
f
or
G
r
e
e
k.
W
he
n
th
e
va
lu
e
s
obt
a
in
e
d
a
r
e
c
om
pa
r
e
d
w
it
h
th
e
va
lu
e
s
of
th
e
s
e
two
s
tu
di
e
s
,
it
i
s
obvi
ous
th
a
t
th
e
r
e
s
ul
t
s
a
r
e
c
lo
s
e
to
t
he
va
lu
e
s
of
th
e
f
ir
s
t
s
tu
dy
a
nd
be
tt
e
r
th
a
n
th
e
s
e
c
ond
s
tu
dy.
T
o
il
lu
s
tr
a
te
th
e
pe
r
f
or
m
a
nc
e
of
th
is
m
ode
l
in
m
or
e
de
ta
il
,
two
e
xa
m
pl
e
s
f
r
om
th
e
da
ta
s
e
t
a
r
e
pr
e
s
e
nt
e
d i
n t
he
T
a
bl
e
3.
F
ig
ur
e
2. T
r
a
in
a
nd va
li
da
ti
on l
os
s
f
or
ne
w
s
da
t
a
s
e
t
w
it
h 8 ba
tc
h
-
s
iz
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
2
,
A
pr
il
20
2
5
:
1587
-
1596
1592
F
ig
ur
e
3. T
r
a
in
a
nd va
li
da
ti
on l
os
s
f
or
ne
w
s
da
t
a
s
e
t
w
it
h 16 ba
t
c
h
-
s
iz
e
T
a
bl
e
1. R
oug
e
m
e
tr
ic
e
s
f
or
T
ur
ki
s
h ne
w
s
da
ta
s
e
t
B
a
t
c
h s
i
z
e
R
ouge
-
1
R
ouge
-
2
R
ouge
-
L
8
31.98
21.11
30.93
16
28.43
17.61
27.40
T
a
bl
e
2. R
oug
e
m
e
tr
ic
e
s
i
n
[
46]
L
a
ngua
ge
R
ouge
-
1
R
ouge
-
2
R
ouge
-
L
S
pa
ni
s
h
30.61
12.36
23.53
C
a
t
a
l
a
n
27.00
11.28
21.27
I
n
th
e
c
ont
in
ua
ti
on
of
th
e
s
tu
dy,
it
w
a
s
a
im
e
d
to
in
c
r
e
a
s
e
th
e
s
uc
c
e
s
s
by
c
ha
ngi
ng
th
e
hype
r
pa
r
a
m
e
te
r
s
.
F
or
th
is
pur
pos
e
,
0.00004
w
a
s
s
e
le
c
te
d
f
or
th
e
le
a
r
ni
ng
r
a
te
a
nd
th
e
s
ys
te
m
w
a
s
r
e
tr
a
in
e
d
on
th
e
ne
w
s
da
ta
s
e
t
a
nd
th
e
r
e
s
ul
t
w
a
s
e
va
lu
a
te
d
w
it
h
R
ouge
m
e
tr
ic
s
a
nd
s
how
n
in
T
a
bl
e
4.
F
a
r
a
ha
ni
e
t
al
.
[
5]
ha
ve
a
c
hi
e
ve
d
42.25,
24.36
,
a
nd
35.94
s
uc
c
e
s
s
e
s
in
th
e
ir
s
tu
dy
f
or
P
e
r
s
ia
n,
r
e
s
p
e
c
ti
ve
ly
.
I
n
a
ddi
ti
on,
B
a
yka
r
a
a
nd
G
üngör
'
s
[
23]
hi
ghe
s
t
va
lu
e
s
obt
a
in
e
d
in
th
e
ir
s
tu
dy
f
or
T
ur
ki
s
h
w
e
r
e
42.26,
27.81
,
a
nd
37.96,
r
e
s
pe
c
ti
ve
ly
.
W
he
n
th
e
v
a
lu
e
s
of
th
is
s
tu
dy
a
r
e
c
om
pa
r
e
d
w
it
h
a
bove
s
tu
di
e
s
,
it
c
a
n
be
s
a
id
th
a
t
th
is
s
tu
dy
a
c
hi
e
ve
d a
good r
e
s
ul
t.
I
n
a
ddi
ti
on,
a
not
he
r
s
tu
dy,
w
hi
c
h
is
in
E
ngl
is
h
N
L
P
s
tu
di
e
s
b
ut
ha
s
not
be
e
n
done
in
T
ur
ki
s
h
unt
il
now
,
w
a
s
pr
oc
e
s
s
e
d
by
th
e
a
ut
hor
a
nd
it
s
e
f
f
e
c
t
on
te
xt
s
um
m
a
r
iz
a
ti
on
w
a
s
e
xa
m
in
e
d.
F
or
th
is
pur
pos
e
,
a
ll
th
e
pl
ur
a
l
s
uf
f
ix
e
s
i
n
t
he
t
e
xt
a
nd
th
e
s
uf
f
ix
e
s
f
ol
lo
w
in
g
t
he
pl
u
r
a
l
s
uf
f
ix
w
e
r
e
de
le
te
d
f
r
om
t
he
w
or
d
a
nd t
hus
th
e
num
be
r
of
w
or
ds
w
a
s
r
e
duc
e
d
a
nd
th
e
e
f
f
e
c
t
on
th
e
te
xt
s
um
m
a
r
iz
a
ti
on
w
a
s
e
xa
m
in
e
d.
A
f
te
r
r
e
m
ovi
ng
th
e
pl
ur
a
l
s
uf
f
ix
a
nd
t
r
a
in
in
g
th
e
s
ys
te
m
w
it
h
th
is
da
ta
s
e
t,
th
e
r
e
s
ul
t
w
a
s
e
va
lu
a
te
d
w
it
h
R
ouge
m
e
tr
ic
s
a
nd
s
how
n
in
T
a
bl
e
5.
S
om
e
ti
m
e
s
m
e
a
ni
ngl
e
s
s
s
um
m
a
r
ie
s
w
e
r
e
pr
oduc
e
d
a
f
te
r
r
e
m
ovi
ng
pl
ur
a
l
s
uf
f
ix
e
s
,
a
nd
th
e
r
e
s
ul
t
w
a
s
a
ls
o v
e
r
y l
ow
c
om
pa
r
e
d t
o ot
he
r
r
e
s
ul
ts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
A
bs
tr
ac
ti
v
e
s
um
m
a
r
iz
at
io
n us
in
g m
ul
ti
li
ngual t
e
x
t
-
to
-
te
x
t
tr
ans
fe
r
t
r
ans
fo
r
m
e
r
f
o
r
t
he
…
(
N
e
da A
li
pour
)
1593
T
a
bl
e
3. E
xa
m
pl
e
s
of
a
b
s
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on f
or
T
ur
ki
s
h ne
w
s
da
ta
s
e
t
E
xa
m
pl
e
M
a
i
n
O
r
i
gi
na
l
a
bs
t
r
a
c
t
m
T
5
a
bs
t
r
a
c
t
E
xa
m
pl
e
1
“
F
e
ne
r
ba
hç
e
kul
übü,
İ
t
a
l
ya
bi
r
i
nc
i
f
ut
bol
l
i
gi
e
ki
pl
e
r
i
nde
n
E
m
pol
i
’
ni
n
o
r
t
a
s
a
ha
oyunc
us
u
M
i
ha
Z
a
j
c
’
ı
ka
dr
os
una
ka
t
t
ı
ğı
nı
r
e
s
m
e
n
a
ç
ı
kl
a
dı
.
S
a
r
ı
-
l
a
c
i
ve
r
t
l
i
kul
übün
i
nt
e
r
ne
t
s
i
t
e
s
i
nde
ye
r
a
l
a
n
a
ç
ı
kl
a
m
a
da
,
24
ya
ş
ı
nda
ki
Z
a
j
c
i
l
e
4,5
yı
l
l
ı
k
a
nl
a
ş
m
a
ya
va
r
ı
l
dı
ğı
be
l
i
r
t
i
l
di
.
Z
a
j
c
t
r
a
ns
f
e
r
i
i
ç
i
n
E
m
pol
i
’
de
ki
r
a
l
ı
k
o
l
a
r
a
k
f
or
m
a
gi
ye
n
S
a
l
i
h
U
ç
a
nı
n
ha
kl
a
r
ı
nda
n
va
z
ge
ç
i
l
di
ği
de
duyu
r
ul
du.
A
ç
ı
kl
a
m
a
da
,
“
K
ul
übüm
üz
,
İ
t
a
l
ya
S
e
r
i
A
e
ki
pl
e
r
i
nde
n
E
m
pol
i
t
a
kı
m
ı
nda
f
o
r
m
a
gi
ye
n
m
e
r
ke
z
or
t
a
s
a
ha
ve
of
a
ns
i
f
or
t
a
s
a
ha
oyunc
us
u
M
i
ha
Z
a
j
c
bons
e
r
vi
s
i
yl
e
bi
r
l
i
kt
e
ka
dr
om
uz
a
k
a
t
m
a
k
üz
e
r
e
kul
übüyl
e
ve
f
ut
bol
c
uyl
a
a
nl
a
ş
m
a
ya
va
r
m
ı
ş
t
ı
r
.
24
ya
ş
ı
nda
ki
S
l
ove
n
oyunc
u
M
i
ha
Z
a
j
c
,
4,5
s
e
z
on
boyun
c
a
s
a
r
ı
-
l
a
c
i
ve
r
t
l
i
f
or
m
a
i
l
e
m
üc
a
de
l
e
e
de
c
e
k.
O
yunc
um
uz
M
i
ha
Z
a
j
c
’
a
F
e
ne
r
ba
hç
e
ye
hoş
ge
l
di
n
di
yor
,
ç
ubukl
u
i
l
e
ni
c
e
ba
ş
a
r
ı
l
a
di
l
i
yor
uz
.
A
yr
ı
c
a
,
bu
t
r
a
ns
f
e
r
ka
ps
a
m
ı
nda
kul
übüm
üz
S
a
l
i
h
U
ç
a
n
üz
e
r
i
nde
ki
ha
kl
a
r
ı
nda
n
va
z
ge
ç
e
r
e
k,
E
m
pol
i
i
l
e
S
a
l
i
h
U
ç
a
n’
ı
n
a
nl
a
ş
m
a
s
ı
na
m
üs
a
a
de
e
t
m
i
ş
t
i
r
”
i
f
a
de
l
e
r
i
ye
r
a
l
dı
.
”
“
F
e
ne
r
ba
hç
e
S
l
ove
n
or
t
a
s
a
ha
oyunc
us
u
M
i
ha
Z
a
j
c
’
ı
4,5
yı
l
l
ı
ğı
na
t
r
a
ns
f
e
r
e
t
t
i
ği
ni
r
e
s
m
e
n
a
ç
ı
kl
a
dı
.
B
u
s
e
z
on
İ
t
a
l
ya
bi
r
i
nc
i
f
ut
bol
l
i
gi
e
ki
bi
E
m
pol
i
’
de
21
r
e
s
m
i
m
a
ç
t
a
gör
e
v
a
l
a
n
Z
a
j
c
,
3
gol
ka
yde
t
t
i
.”
“
F
e
ne
r
ba
hç
e
kul
übü,
E
m
pol
i
’
ni
n or
t
a
s
a
ha
oyunc
us
u
M
i
ha
Z
a
j
c
’
ı
ka
dr
os
una
ka
t
t
ı
ğı
nı
r
e
s
m
e
a
ç
ı
kl
a
dı
.”
E
xa
m
pl
e
2
“
M
uğl
a
’
nı
n
B
odr
um
i
l
ç
e
s
i
nde
,
i
ç
e
r
i
s
i
nde
a
s
ke
r
i
pe
r
s
on
e
l
i
n
bul
unduğu m
i
ni
büs
s
u ka
n
a
l
ı
na
düş
t
ü.
K
a
z
a
da
m
i
ni
büs
t
e
bul
una
n 3
a
s
ke
r
ha
f
i
f
ş
e
ki
l
de
y
a
r
a
l
a
ndı
.
K
a
z
a
,
bugün
a
kş
a
m
s
a
a
t
l
e
r
i
nde
B
odr
um
-
M
i
l
a
s
ka
r
a
yol
u
üz
e
r
i
nde
m
e
yda
na
ge
l
di
.
A
s
ke
r
i
pe
r
s
one
l
t
a
ş
ı
ya
n
48
T
N
173
pl
a
ka
l
ı
m
i
ni
büs
,
G
üve
r
c
i
nl
i
k
i
s
t
i
ka
m
e
t
i
ne
gi
de
r
ke
n,
s
a
ğa
na
k
ya
ğı
ş
s
onr
a
s
ı
ka
yga
nl
a
ş
a
n
yol
da
s
ür
üc
ü
di
r
e
ks
i
yon
ha
ki
m
i
ye
t
i
ni
ka
ybe
t
t
i
.
K
ont
r
ol
de
n
ç
ı
ka
n
m
i
ni
büs
,
önc
e
yol
ke
na
r
ı
nda
bul
una
n
s
u ka
na
l
ı
na
düş
t
ü, d
a
ha
s
onr
a
da
ka
y
a
l
ı
kl
a
r
a
ç
a
r
pa
r
a
k
dur
a
bi
l
di
.
K
a
z
a
nı
n
a
r
dı
nda
n
ol
a
y
ye
r
i
ne
ge
l
e
n
M
uğl
a
911
A
r
a
m
a
K
ur
t
a
r
m
a
e
ki
pl
e
r
i
ha
f
i
f
ya
r
a
l
ı
a
s
k
e
r
l
e
r
i
a
r
a
ş
t
a
n
ç
ı
ka
r
a
r
a
k
s
a
ğl
ı
k
e
ki
pl
e
r
i
ne
t
e
s
l
i
m
e
t
t
i
.
Y
a
r
a
l
a
n
3
a
s
ke
r
de
n
2’
s
i
B
odr
um
D
e
vl
e
t
H
a
s
t
a
ne
s
i
’
ne
,
1
a
s
ke
r
i
s
e
öz
e
l
bi
r
ha
s
t
a
ne
ye
ka
l
dı
r
ı
l
dı
.
T
e
da
vi
ye
a
l
ı
na
n 3 a
s
ke
r
i
n de
s
a
ğl
ı
k dur
um
u i
yi
ol
duğu öğr
e
ni
l
di
.”
“
M
uğl
a
’
nı
n
B
odr
um
i
l
ç
e
s
i
nde
a
s
k
e
r
i
pe
r
s
one
l
t
a
ş
ı
ya
n
a
s
ke
r
i
m
i
ni
büs
ka
z
a
ya
pt
ı
.
Y
ol
ke
na
r
ı
nda
bul
una
n
s
u
ka
na
l
ı
na
düş
e
n
m
i
ni
büs
t
e
bul
una
n
3
a
s
ke
r
ha
f
i
f
ş
e
ki
l
de
ya
r
a
l
a
ndı
.”
“
M
uğl
a
’
nı
n
B
odr
um
i
l
ç
e
s
i
nde
,
i
ç
e
r
i
s
i
nde
a
s
ke
r
i
pe
r
s
one
l
i
n
bul
unduğu
m
i
ni
büs
s
u
ka
na
l
ı
na
dü
ş
t
ü.
K
a
z
a
da
3
a
s
ke
r
ya
r
a
l
a
ndı
.”
T
a
bl
e
4. R
oug
e
m
e
tr
ic
e
s
f
or
T
ur
ki
s
h ne
w
s
da
ta
s
e
t
w
it
h c
ha
ngi
n
g l
e
a
r
ni
ng r
a
te
B
a
t
c
h s
i
z
e
L
e
a
r
ni
ng r
a
t
e
R
ouge
-
1
R
ouge
-
2
R
ouge
-
L
8
0
.
00004
58
.
76
52
.
98
58
.
45
T
a
bl
e
5. R
oug
e
m
e
tr
ic
e
s
f
or
T
ur
ki
s
h ne
w
s
da
ta
s
e
t
by r
e
m
ovi
ng plur
a
l
s
uf
f
ix
e
s
B
a
t
c
h s
i
z
e
R
ouge
-
1
R
ouge
-
2
R
ouge
-
L
8
10
.
55
3
.
89
10
.
21
A
f
te
r
th
e
ne
w
s
da
ta
s
e
t,
th
e
m
ode
l
w
a
s
te
s
t
e
d
on
th
e
pa
pe
r
da
t
a
s
e
t.
H
ow
e
v
e
r
,
s
in
c
e
th
e
p
a
pe
r
da
ta
s
e
t
w
a
s
la
r
ge
,
th
e
pa
pe
r
s
w
e
r
e
f
ir
s
t
r
e
duc
e
d
to
26
li
ne
s
w
it
h
t
he
B
E
R
T
e
xt
r
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on
m
e
th
od.
B
e
c
a
us
e
th
e
s
m
a
ll
e
s
t
pa
pe
r
ha
d
26
li
ne
s
.
T
hus
,
th
e
s
iz
e
of
th
e
pa
pe
r
da
ta
s
e
t
w
a
s
r
e
duc
e
d
a
nd
gi
ve
n
to
th
e
s
ys
te
m
. T
he
p
a
pe
r
da
ta
w
e
r
e
t
r
a
in
e
d w
it
h 8 ba
tc
h s
iz
e
a
nd 15 e
poc
hs
. be
c
a
u
s
e
, be
tt
e
r
r
e
s
ul
t
w
a
s
obt
a
in
e
d w
it
h
th
is
ba
tc
h
s
iz
e
on
T
ur
ki
s
h
ne
w
s
.
T
r
a
in
a
nd
va
li
da
ti
on
lo
s
s
f
or
pa
pe
r
s
w
it
h
8
ba
tc
h
s
iz
e
w
e
r
e
s
how
n
in
F
ig
ur
e
4.
A
f
te
r
tr
a
in
in
g
th
e
m
ode
l
w
it
h
ba
tc
h
s
iz
e
s
of
8
a
nd
15
e
poc
h
s
,
th
e
m
o
s
t
c
om
m
onl
y
us
e
d
R
oug
e
m
e
tr
ic
s
w
e
r
e
us
e
d
a
nd
th
e
s
uc
c
e
s
s
of
th
e
m
ode
l
w
a
s
e
xa
m
in
e
d
by
obt
a
in
in
g
R
ouge
1,
R
oug
e
2
,
a
nd R
ouge
L
.
R
ouge
-
1, R
ouge
-
2, a
nd R
ouge
-
L
va
lu
e
s
w
e
r
e
s
how
n i
n
T
a
bl
e
6.
T
a
bl
e
6. R
oug
e
m
e
tr
ic
e
s
f
or
pa
pe
r
s
da
ta
s
e
t
B
a
t
c
h s
i
z
e
R
ouge
-
1
R
ouge
-
2
R
ouge
-
L
8
18.34
4.62
17.63
U
nt
il
now
, no a
bs
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on pr
oc
e
s
s
w
it
h m
T
5 ha
s
be
e
n pe
r
f
or
m
e
d
on t
he
pa
pe
r
da
ta
s
e
t.
I
n
a
ddi
ti
on,
it
is
not
pos
s
ib
le
to
c
om
p
a
r
e
th
e
r
e
s
ul
ts
of
th
i
s
s
tu
dy
w
it
h
ot
he
r
s
tu
di
e
s
,
be
c
a
us
e
T
ur
ki
s
h
a
r
ti
c
le
da
ta
s
e
t
w
a
s
c
r
e
a
te
d by the
a
ut
hor
. A
ls
o, t
he
r
e
a
s
on f
or
t
he
l
ow
r
ouge
m
e
tr
ic
s
i
s
t
ha
t
th
e
t
e
xt
i
s
m
e
a
ni
ngf
ul
but
lo
ng.
S
o,
th
e
pr
oduc
e
d
te
xt
s
m
a
y
di
f
f
e
r
f
r
om
th
e
a
c
tu
a
l
a
bs
tr
a
c
t
s
.
T
h
e
r
e
f
or
e
,
th
e
r
e
s
ul
ts
of
th
is
s
tu
dy
c
a
n
f
or
m
a
ba
s
is
f
or
f
ut
ur
e
s
um
m
a
r
iz
a
ti
on
s
tu
di
e
s
.
A
n
e
xa
m
pl
e
f
r
om
th
e
da
ta
s
e
t
is
pr
e
s
e
nt
e
d
in
th
e
ta
bl
e
to
il
lu
s
tr
a
te
th
e
pe
r
f
or
m
a
nc
e
of
th
is
m
ode
l
in
m
o
r
e
de
ta
il
.
S
in
c
e
th
e
te
xt
is
lo
ng,
onl
y
th
e
s
um
m
a
r
y
a
nd
th
e
s
um
m
a
r
y
pr
oduc
e
d w
it
h t
he
m
ode
l
a
r
e
pr
e
s
e
nt
e
d i
n
T
a
bl
e
7.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
2
,
A
pr
il
20
2
5
:
1587
-
1596
1594
F
ig
ur
e
4. T
r
a
in
a
nd va
li
da
ti
on
l
os
s
f
or
pa
pe
r
s
da
ta
s
e
t
T
a
bl
e
7. E
xa
m
pl
e
s
of
a
b
s
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on f
or
pa
pe
r
s
da
ta
s
e
t
O
r
i
gi
na
l
A
bs
t
r
a
c
t
m
T
5 A
bs
t
r
a
c
t
“
i
ki
nc
i
dünya
s
a
va
ş
ı
s
onr
a
s
ı
ül
ke
l
e
r
a
r
a
s
ı
ge
l
i
ş
m
i
ş
l
i
k
f
a
r
kl
a
r
ı
nı
n
be
l
i
r
gi
nl
e
ş
m
e
s
i
yl
e
a
z
ge
l
i
ş
m
i
ş
ve
y
a
ge
r
i
ka
l
m
ı
ş
ül
ke
l
e
r
i
n
e
konom
i
k
ol
a
r
a
k
ka
l
kı
nm
a
s
ı
s
on
d
e
r
e
c
e
c
i
ddi
bi
r
s
or
un
ol
a
r
a
k
or
t
a
ya
ç
ı
km
ı
ş
t
ı
r
.
bu
s
ür
e
ç
t
e
ge
l
i
ş
m
i
ş
l
i
k
f
a
r
l
ı
l
ı
kl
a
r
ı
ba
kı
m
ı
nda
n
be
nz
e
r
ol
m
a
ya
n
ül
ke
l
e
r
i
n
be
nz
e
r
büyüm
e
m
ode
l
l
e
r
i
ni
uygul
a
m
a
l
a
r
ı
nı
n
m
üm
kün
ol
m
a
dı
ğı
gö
r
ül
m
üş
ve
ka
l
kı
nm
a
ç
a
ba
l
a
r
ı
nda
ye
ni
a
r
a
yı
ş
l
a
r
a
yöne
l
m
i
ş
t
i
r
.
bu
döne
m
de
ol
uş
m
a
ya
ba
ş
l
a
ya
n
büyüm
e
t
e
or
i
l
e
r
i
ni
n
öz
ünü
s
a
va
ş
s
onr
a
s
ı
s
a
va
ş
t
a
n
e
t
ki
l
e
ne
n
e
konom
i
l
e
r
i
n
ka
l
kı
ndı
r
ı
l
m
a
s
ı
o
l
uş
t
ur
m
uş
t
ur
.
bu
doğr
ul
t
uda
ge
l
i
ş
e
n
büyüm
e
t
e
or
i
l
e
r
i
ül
ke
l
e
r
i
n
ge
l
i
ş
m
e
ç
a
ba
l
a
r
ı
nda
öne
m
l
i
r
ol
oyna
m
ı
ş
t
ı
r
.
büyüm
e
ni
n
t
e
m
e
l
be
l
i
r
l
e
yi
c
i
l
e
r
i
üz
e
r
i
nde
ya
pı
l
a
n
de
ğe
r
l
e
ndi
r
m
e
l
e
r
i
l
e
ge
l
i
ş
m
e
yol
unda
i
vm
e
ka
z
a
nı
l
m
ı
ş
t
ı
r
.
a
nc
a
k
ül
ke
l
e
r
i
n
ka
l
kı
nm
a
l
a
r
ı
nda
t
e
m
e
l
be
l
i
r
l
e
yi
c
i
l
e
r
de
n
ol
a
n
e
ği
t
i
m
f
a
kt
ör
ü
be
ş
e
r
i
s
e
r
m
a
ye
ol
uş
um
una
ka
t
kı
s
a
ğl
a
ya
r
a
k
i
kt
i
s
a
di
büyüm
e
de
ön
e
m
l
i
ol
m
a
kt
a
dı
r
.
e
ği
t
i
m
be
ş
e
r
i
s
e
r
m
a
ye
t
e
or
i
s
i
ni
n
k
i
l
i
t
uns
ur
l
a
r
ı
nda
n
bi
r
i
di
r
ç
ünkü
bi
l
gi
ve
be
c
e
r
i
yi
ge
l
i
ş
t
i
r
m
e
ni
n
bi
r
i
nc
i
l
yol
u
ol
a
r
a
k
gör
ül
m
e
kt
e
di
r
.
buna
gör
e
e
ği
t
i
m
düz
e
yi
e
m
e
k
ka
l
i
t
e
s
i
ni
öl
ç
m
e
ni
n
bi
r
yol
u
ol
a
r
a
k
e
l
e
a
l
ı
nm
a
kt
a
dı
r
.
ni
t
e
l
i
kl
i
e
ği
t
i
m
i
s
e
be
ş
e
r
i
s
e
r
m
a
ye
ol
u
ş
um
unun
t
e
m
e
l
i
ni
ol
uş
t
ur
m
a
kt
a
dı
r
. büyüm
e
t
e
or
i
l
e
r
i
ni
n ge
l
i
ş
i
m
i
.”
“
bu
ç
a
l
ı
ş
m
a
d
a
t
ür
ki
ye
de
m
e
s
l
e
ki
ve
t
e
kni
k
e
ği
t
i
m
kur
um
l
a
r
ı
i
t
i
ba
r
i
yl
e
nüf
us
be
ş
e
r
i
s
e
r
m
a
ye
ni
n
ge
l
i
ş
m
e
s
i
yl
e
bi
r
l
i
kt
e
yür
üt
ül
m
üş
t
ür
.
bu
ka
ps
a
m
da
t
ür
ki
ye
de
m
e
s
l
e
ki
ve
t
e
kni
k
e
ği
t
i
m
kur
um
l
a
r
ı
i
t
i
ba
r
i
yl
e
e
konom
i
k
büyüm
e
m
ode
l
l
e
r
i
ni
n
or
t
a
ya
ç
ı
kt
ı
ğı
bi
r
döne
m
di
r
.
t
ü
r
ki
ye
de
m
e
s
l
e
ki
ve
t
e
kni
k
e
ği
t
i
m
kur
um
l
a
r
ı
i
t
i
ba
r
i
yl
e
i
ns
a
n
s
e
r
m
a
ye
s
i
i
l
e
s
a
ğl
a
nm
ı
ş
t
ı
r
.
bu
döne
m
de
e
konom
i
ni
n
s
ür
dür
ül
e
bi
l
i
r
l
i
ği
ni
n
a
r
t
m
a
s
ı
na
yöne
l
i
k
s
onuç
l
a
r
or
t
a
ya
ç
ı
ka
n
ni
t
e
l
i
kl
i
i
ş
güc
ü
i
ht
i
ya
c
ı
ka
r
ş
ı
l
a
m
a
kt
a
dı
r
.
bu
dur
um
un
s
onunda
be
ş
e
r
i
s
e
r
m
a
ye
ni
n
ge
l
i
ş
m
e
s
i
ne
ka
t
kı
s
a
ğl
a
dı
ğı
düş
ünül
m
e
kt
e
di
r
.
”
5.
C
O
N
C
L
U
S
I
O
N
T
he
r
e
a
r
e
a
li
m
it
e
d
num
be
r
of
s
tu
di
e
s
on
a
b
s
tr
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on
w
it
h
pr
e
-
tr
a
in
e
d
m
ode
ls
in
T
ur
ki
s
h
te
xt
s
.
I
n
th
is
s
tu
dy,
a
pr
e
-
tr
a
in
e
d
m
T
5
m
ode
l
w
a
s
u
s
e
d
to
s
um
m
a
r
iz
e
T
ur
ki
s
h
te
xt
s
.
T
hi
s
m
ode
l
w
a
s
f
ir
s
t
te
s
te
d
on
M
L
S
U
M
da
t
a
s
e
t
w
hi
c
h
is
ne
w
s
da
ta
s
e
t,
a
nd
th
e
n
on
th
e
a
r
ti
c
le
da
ta
s
e
t
c
r
e
a
te
d
by
th
e
a
ut
hor
.
S
in
c
e
th
e
a
r
ti
c
le
s
w
e
r
e
lo
ng,
th
e
ir
s
e
nt
e
n
c
e
s
w
e
r
e
r
e
duc
e
d
by
t
he
B
E
R
T
e
xt
r
a
c
ti
ve
s
um
m
a
r
iz
a
ti
on.
D
ue
to
a
la
c
k
of
s
tu
di
e
s
a
nd
da
ta
s
e
t
in
a
r
ti
c
le
da
ta
s
e
t,
pr
e
pa
r
in
g
th
e
da
ta
s
e
t
w
a
s
one
of
th
e
m
os
t
im
por
ta
nt
li
m
it
a
ti
ons
a
nd
di
f
f
ic
ul
ti
e
s
of
th
e
s
tu
dy
a
nd
our
p
a
pe
r
c
oul
d
not
be
c
om
pa
r
e
d
to
a
ny
e
a
r
li
e
r
s
tu
dy
a
nd
c
a
n
now
s
e
r
ve
a
s
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
A
bs
tr
ac
ti
v
e
s
um
m
a
r
iz
at
io
n us
in
g m
ul
ti
li
ngual t
e
x
t
-
to
-
te
x
t
tr
ans
fe
r
t
r
ans
fo
r
m
e
r
f
o
r
t
he
…
(
N
e
da A
li
pour
)
1595
ba
s
e
li
ne
f
or
a
ny
f
ut
ur
e
s
tu
di
e
s
in
th
is
f
ie
ld
.
A
not
he
r
m
os
t
i
m
por
ta
nt
li
m
it
a
ti
on
is
th
e
s
ys
te
m
in
a
de
qua
c
y.
T
e
xt
da
ta
s
e
t
s
r
e
qui
r
e
hi
gh
ha
r
dw
a
r
e
,
a
nd
a
lt
hough
th
is
w
or
k
r
a
n
on
G
oogl
e
C
ol
a
b
P
r
o
P
lu
s
,
it
w
a
s
m
e
t
w
it
h
ha
r
dw
a
r
e
f
a
il
ur
e
e
r
r
or
s
in
m
os
t
c
a
s
e
s
.
I
n
a
ddi
ti
on,
T
ur
ki
s
h
la
ngua
ge
is
a
n
a
ggl
ut
in
a
ti
ve
la
ngua
ge
a
nd
N
L
P
s
tu
di
e
s
a
r
e
v
e
r
y
di
f
f
ic
ul
t
in
T
ur
ki
s
h.
I
n
f
ut
ur
e
s
tu
di
e
s
,
it
c
a
n
be
e
xa
m
in
e
d
w
he
th
e
r
s
uc
c
e
s
s
h
a
s
c
ha
nge
d
by
e
nl
a
r
gi
ng
th
e
a
r
ti
c
le
d
a
ta
s
e
t.
I
n
a
ddi
ti
on,
th
e
s
uc
c
e
s
s
of
th
e
m
o
de
l
c
a
n
b
e
e
v
a
lu
a
te
d
by
e
nr
ic
hi
ng
th
e
s
y
s
te
m
ha
r
dw
a
r
e
,
c
ha
ngi
ng
th
e
hype
r
pa
r
a
m
e
te
r
s
a
nd
doubli
ng
th
e
da
ta
s
e
t.
O
n
th
e
ot
he
r
ha
nd,
in
a
ddi
t
io
n
to
te
xt
s
um
m
a
r
iz
a
ti
on, que
s
ti
on ge
ne
r
a
to
r
c
a
n be
i
m
por
ta
nt
r
e
s
e
a
r
c
h.
R
E
F
E
R
E
N
C
E
S
[
1]
A
.
N
e
nkova
a
nd
K
.
M
c
K
e
ow
n,
“
A
s
ur
ve
y
of
t
e
xt
s
um
m
a
r
i
z
a
t
i
on
t
e
c
hni
que
s
,
”
i
n
M
i
ni
ng
T
e
x
t
D
at
a
,
B
os
t
on,
M
A
:
S
pr
i
nge
r
U
S
,
2012, pp. 43
–
76
, doi
:
10.1007/
978
-
1
-
4614
-
3223
-
4_3.
[
2]
H
.
P
.
E
dm
unds
on,
“
N
e
w
m
e
t
hods
i
n
a
ut
om
a
t
i
c
e
xt
r
a
c
t
i
ng,”
J
our
nal
of
t
he
A
C
M
,
vol
.
16,
no.
2,
pp.
264
–
285,
1969,
doi
:
10.1145/
321510.321519.
[
3]
A
.
T
ur
pi
n,
Y
.
T
s
e
ga
y,
D
.
H
a
w
ki
ng,
a
nd
H
.
E
.
W
i
l
l
i
a
m
s
,
“
F
a
s
t
g
e
ne
r
a
t
i
on
of
r
e
s
ul
t
s
ni
ppe
t
s
i
n
w
e
b
s
e
a
r
c
h,”
i
n
P
r
oc
e
e
di
ngs
of
t
h
e
30t
h annual
i
nt
e
r
nat
i
onal
A
C
M
SI
G
I
R
c
onf
e
r
e
nc
e
on R
e
s
e
ar
c
h and de
v
e
l
opm
e
nt
i
n i
nf
or
m
at
i
on r
e
t
r
i
e
v
al
, N
e
w
Y
or
k, U
S
A
:
A
C
M
,
2007, pp. 127
–
134
, doi
:
10.1145/
1277741.1277766.
[
4]
J
. M
. T
. M
or
e
no,
A
ut
om
at
i
c
t
e
x
t
s
um
m
ar
i
z
at
i
on
. H
oboke
n, N
e
w
J
e
r
s
e
y:
J
ohn W
i
l
e
y &
S
ons
, 2014
, doi
:
10.1002/
9781119004752.
[
5]
M
.
F
a
r
a
ha
ni
,
M
.
G
ha
r
a
c
hor
l
oo,
a
nd
M
.
M
a
nt
hour
i
,
“
L
e
ve
r
a
gi
ng
pa
r
s
b
e
r
t
a
nd
pr
e
t
r
a
i
ne
d
m
T
5
f
or
pe
r
s
i
a
n
a
bs
t
r
a
c
t
i
ve
t
e
xt
s
um
m
a
r
i
z
a
t
i
on,”
i
n
26t
h
I
nt
e
r
nat
i
onal
C
om
put
e
r
C
onf
e
r
e
nc
e
,
C
om
put
e
r
So
c
i
e
t
y
of
I
r
an,
C
SI
C
C
2021
,
I
E
E
E
,
2021,
pp.
1
–
6
,
doi
:
10.1109/
C
S
I
C
C
52343.2021.9420563.
[
6]
J
.
C
hr
i
s
t
e
ns
e
n,
M
a
us
a
m
,
S
.
S
ode
r
l
a
nd,
a
nd
O
.
E
t
z
i
oni
,
“
T
ow
a
r
ds
c
ohe
r
e
nt
m
ul
t
i
-
doc
um
e
nt
s
um
m
a
r
i
z
a
t
i
on,”
i
n
P
r
oc
e
e
di
ngs
of
t
he
2013
C
onf
e
r
e
nc
e
of
t
he
N
or
t
h
A
m
e
r
i
c
an
C
hapt
e
r
of
t
he
A
s
s
oc
i
at
i
on
f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
:
H
um
an
L
anguage
T
e
c
hnol
ogi
e
s
, A
s
s
oc
i
a
t
i
on f
or
C
om
put
a
t
i
ona
l
L
i
ngui
s
t
i
c
s
, 2013, pp. 1163
–
1173
.
[
7]
A
.
N
e
nkova
,
L
.
V
a
nde
r
w
e
nde
,
a
nd
K
.
M
c
K
e
ow
n,
“
A
c
om
pos
i
t
i
ona
l
c
o
nt
e
xt
s
e
ns
i
t
i
ve
m
ul
t
i
-
doc
um
e
nt
s
um
m
a
r
i
z
e
r
,”
i
n
P
r
oc
e
e
di
ngs
of
t
he
29t
h annual
i
nt
e
r
nat
i
onal
A
C
M
SI
G
I
R
c
onf
e
r
e
nc
e
on R
e
s
e
a
r
c
h and de
v
e
l
opm
e
nt
i
n i
nf
o
r
m
at
i
on r
e
t
r
i
e
v
al
, N
e
w
Y
or
k, U
S
A
:
A
C
M
, 2006, pp. 573
–
580
, doi
:
10.1145/
1148170.1148269.
[
8]
M
.
G
a
m
bhi
r
a
nd
V
.
G
upt
a
,
“
R
e
c
e
nt
a
ut
om
a
t
i
c
t
e
xt
s
um
m
a
r
i
z
a
t
i
on
t
e
c
hni
que
s
:
a
s
ur
ve
y,”
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
R
e
v
i
e
w
,
vol
.
47
,
no. 1, pp. 1
–
66, 2017, doi
:
10.1007/
s
10462
-
016
-
9475
-
9.
[
9]
V
.
G
upt
a
a
nd
G
.
S
.
L
e
ha
l
,
“
A
s
ur
ve
y
of
t
e
xt
s
um
m
a
r
i
z
a
t
i
on
e
xt
r
a
c
t
i
ve
t
e
c
hni
que
s
,”
J
ou
r
nal
of
E
m
e
r
gi
ng
T
e
c
hnol
ogi
e
s
i
n
W
e
b
I
nt
e
l
l
i
ge
nc
e
, vol
. 2, no. 3, pp. 258
–
268, 2010, doi
:
10.4304/
j
e
t
w
i
.2.3.258
-
268.
[
10]
S
.
C
hopr
a
,
M
.
A
ul
i
,
a
nd
A
.
M
.
R
us
h,
“
A
bs
t
r
a
c
t
i
ve
s
e
nt
e
nc
e
s
um
m
a
r
i
z
a
t
i
on
w
i
t
h
a
t
t
e
nt
i
ve
r
e
c
ur
r
e
nt
ne
ur
a
l
ne
t
w
or
ks
,”
i
n
2016
C
onf
e
r
e
nc
e
of
t
he
N
or
t
h
A
m
e
r
i
c
an
C
hapt
e
r
of
t
he
A
s
s
oc
i
at
i
on
f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
:
H
um
an
L
anguage
T
e
c
hnol
ogi
e
s
,
2016, pp. 93
–
98
, doi
:
10.18653/
v1/
n16
-
1012.
[
11]
R
.
N
a
l
l
a
pa
t
i
,
B
.
Z
hou,
C
.
D.
S
a
nt
o
s
,
Ç
.
G
ul
ç
e
hr
e
,
a
nd
B
.
X
i
a
ng,
“
A
bs
t
r
a
c
t
i
v
e
t
e
xt
s
um
m
a
r
i
z
a
t
i
on
us
i
ng
s
e
que
nc
e
-
to
-
s
e
que
nc
e
R
N
N
s
a
nd
be
yond,
”
i
n
C
oN
L
L
2016
-
20t
h
SI
G
N
L
L
C
onf
e
r
e
nc
e
on
C
om
put
at
i
onal
N
at
ur
al
L
anguage
L
e
ar
ni
ng,
P
r
oc
e
e
di
ng
s
,
F
e
b. 2016, pp. 280
–
290
, doi
:
10.18653/
v1/
k16
-
1028.
[
12]
S
.
H
oc
hr
e
i
t
e
r
a
nd
J
.
S
c
hm
i
dhube
r
,
“
L
ong
s
hor
t
-
t
e
r
m
m
e
m
or
y,”
N
e
ur
al
C
om
put
at
i
on
,
vol
.
9,
no.
8,
pp.
1735
–
1780,
1997,
doi
:
10.1162/
ne
c
o.1997.9.8.1735.
[
13]
A
. V
a
s
w
a
ni
e
t
al
.
, “
A
t
t
e
nt
i
on i
s
a
l
l
you ne
e
d,”
i
n
31s
t
C
onf
e
r
e
n
c
e
on N
e
u
r
al
I
nf
or
m
at
i
on P
r
oc
e
s
s
i
ng Sy
s
t
e
m
s
, 2017, pp. 1
–
11.
[
14]
L
.
D
ong
e
t
al
.
,
“
U
ni
f
i
e
d
l
a
ngua
ge
m
ode
l
pr
e
-
t
r
a
i
ni
ng
f
or
na
t
ur
a
l
l
a
ngua
ge
u
nde
r
s
t
a
ndi
ng
a
nd
ge
ne
r
a
t
i
on,”
A
dv
anc
e
s
i
n
N
e
ur
a
l
I
nf
or
m
at
i
on P
r
oc
e
s
s
i
ng Sy
s
t
e
m
s
, vol
. 32, 2019, doi
:
10.48550/
a
r
X
i
v.1905.0319
7.
[
15]
K
.
S
ong,
X
.
T
a
n,
T
.
Q
i
n,
J
.
L
u,
a
nd
T
.
Y
.
L
i
u,
“
M
A
S
S
:
m
a
s
ke
d
s
e
que
n
c
e
t
o
s
e
que
nc
e
pr
e
-
t
r
a
i
ni
ng
f
or
l
a
ngua
ge
ge
n
e
r
a
t
i
on,”
i
n
36t
h I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on M
ac
hi
ne
L
e
a
r
ni
ng, I
C
M
L
2019
, C
a
l
i
f
or
ni
a
:
P
M
L
R
97, 2019, pp. 10384
–
10394.
[
16]
S
.
R
ot
he
,
S
.
N
a
r
a
ya
n,
a
nd
A
.
S
e
ve
r
yn,
“
L
e
ve
r
a
gi
ng
pr
e
-
t
r
a
i
ne
d
c
he
c
kpoi
nt
s
f
or
s
e
que
nc
e
ge
ne
r
a
t
i
on
t
a
s
ks
,”
T
r
ans
a
c
t
i
ons
of
t
he
A
s
s
oc
i
at
i
on f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
, vol
. 8, pp. 264
–
280, 2020, doi
:
10.1
162/
t
a
c
l
_a
_00313.
[
17]
M
.
L
e
w
i
s
e
t
al
.
,
“
B
A
R
T
:
de
noi
s
i
ng
s
e
que
nc
e
-
to
-
s
e
que
nc
e
pr
e
-
t
r
a
i
ni
ng
f
or
na
t
ur
a
l
l
a
ngua
ge
ge
ne
r
a
t
i
on,
t
r
a
ns
l
a
t
i
on,
a
nd
c
om
pr
e
he
ns
i
on,”
i
n
P
r
oc
e
e
di
ngs
of
t
he
58t
h A
nnual
M
e
e
t
i
ng of
t
he
A
s
s
oc
i
at
i
on
f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
,
S
t
r
ouds
bur
g, U
S
A
:
A
s
s
oc
i
a
t
i
on f
or
C
om
put
a
t
i
ona
l
L
i
ngui
s
t
i
c
s
, 2020, pp. 7871
–
7880
, doi
:
10.1865
3/
v1/
2020.a
c
l
-
m
a
i
n.703.
[
18]
J
. Z
ha
ng, Y
. Z
ha
o, M
. S
a
l
e
h, a
nd P
. J
. L
i
u, “
P
E
G
A
S
U
S
:
pr
e
-
t
r
a
i
ni
ng w
i
t
h e
xt
r
a
c
t
e
d ga
p
-
s
e
nt
e
nc
e
s
f
or
a
bs
t
r
a
c
t
i
ve
s
um
m
a
r
i
z
a
t
i
on,”
i
n
37t
h I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on M
ac
hi
ne
L
e
a
r
ni
ng, I
C
M
L
2020
, P
M
L
R
11
9, 2020, pp. 11265
–
11276.
[
19]
J
.
D
e
vl
i
n,
M
.
-
W
.
C
ha
ng,
K
.
L
e
e
,
K
.
T
.
G
oogl
e
,
a
nd
A
.
I
.
L
a
ngua
g
e
,
“
B
E
R
T
:
pr
e
-
t
r
a
i
ni
ng
of
de
e
p
bi
di
r
e
c
t
i
ona
l
t
r
a
ns
f
or
m
e
r
s
f
or
l
a
ngua
ge
unde
r
s
t
a
ndi
ng,”
i
n
P
r
oc
e
e
di
ngs
of
N
A
A
C
L
-
H
L
T
2019
, 2019, pp. 4171
–
4186.
[
20]
L
. X
ue
e
t
al
.
, “
M
T
5:
a
m
a
s
s
i
ve
l
y m
ul
t
i
l
i
ngua
l
pr
e
-
t
r
a
i
ne
d t
e
xt
-
to
-
t
e
xt
t
r
a
ns
f
or
m
e
r
,”
ar
X
i
v
-
C
om
put
e
r
Sc
i
e
nc
e
,
pp. 1
-
17,
O
c
t
. 2020.
[
21]
Y
.
L
i
u
e
t
al
.
,
“
M
ul
t
i
l
i
ngua
l
de
noi
s
i
ng
pr
e
-
t
r
a
i
ni
ng
f
o
r
ne
ur
a
l
m
a
c
hi
ne
t
r
a
ns
l
a
t
i
on,”
T
r
ans
ac
t
i
on
s
of
t
he
A
s
s
oc
i
at
i
on
f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
, vol
. 8, pp. 726
–
742, 2020, doi
:
10.1162/
t
a
c
l
_a
_00343.
[
22]
T
.
S
c
i
a
l
om
,
P
.
A
.
D
r
a
y,
S
.
L
a
m
pr
i
e
r
,
B
.
P
i
w
ow
a
r
s
ki
,
a
nd
J
.
S
t
a
i
a
no,
“
M
L
S
U
M
:
t
he
m
ul
t
i
l
i
ngua
l
s
um
m
a
r
i
z
a
t
i
on
c
or
pus
,”
i
n
E
M
N
L
P
2020
-
2020
C
onf
e
r
e
nc
e
on
E
m
pi
r
i
c
al
M
e
t
hods
i
n
N
at
ur
al
L
anguage
P
r
oc
e
s
s
i
ng,
P
r
oc
e
e
di
ngs
of
t
he
C
onf
e
r
e
n
c
e
,
2020,
pp. 8051
–
8067
, doi
:
10.18653/
v1/
2020.e
m
nl
p
-
m
a
i
n.647.
[
23]
B
.
B
a
yka
r
a
a
nd
T
.
G
üngör
,
“
T
ur
ki
s
h
a
bs
t
r
a
c
t
i
ve
t
e
xt
s
um
m
a
r
i
z
a
t
i
on
u
s
i
ng
pr
e
t
r
a
i
ne
d
s
e
que
nc
e
-
to
-
s
e
que
nc
e
m
ode
l
s
,
”
N
at
u
r
al
L
anguage
E
ngi
ne
e
r
i
ng
, vol
. 29, no. 5, pp. 1275
–
1304, 2023, doi
:
10.1017/
S
1351324922000195.
[
24]
G
.
E
r
yi
ği
t
,
J
.
N
i
vr
e
,
a
nd
K
.
O
f
l
a
z
e
r
,
“
D
e
pe
nde
nc
y
pa
r
s
i
ng
of
t
ur
ki
s
h,”
C
om
p
ut
at
i
onal
L
i
ngui
s
t
i
c
s
,
vol
.
34,
no.
4,
pp.
357
–
389,
2008, doi
:
10.1162/
c
ol
i
.2008.34.4.627.
[
25]
D
.
Z
.
H
a
kka
ni
-
T
ür
,
K
.
O
f
l
a
z
e
r
,
a
nd
G
.
T
ür
,
“
S
t
a
t
i
s
t
i
c
a
l
m
or
phol
ogi
c
a
l
di
s
a
m
bi
gua
t
i
on
f
or
a
ggl
ut
i
na
t
i
ve
l
a
ngua
ge
s
,”
C
om
put
e
r
s
and t
he
H
um
ani
t
i
e
s
, vol
. 36, no. 4, pp. 381
–
410, 2002, doi
:
10.1023/
A
:
1020271707826.
[
26]
A
. R
a
df
or
d, K
. N
a
r
a
s
i
m
ha
n, T
. S
a
l
i
m
a
ns
,
a
nd I
. S
ut
s
ke
ve
r
, “
I
m
pr
ovi
ng l
a
ngua
g
e
unde
r
s
t
a
ndi
ng w
i
t
h uns
up
e
r
vi
s
e
d l
e
a
r
ni
ng,”
O
pe
n
A
I
,
pp. 1
-
12,
2018.
[
27]
Z
.
Y
a
ng,
Z
.
D
a
i
,
Y
.
Y
a
ng,
J
.
C
a
r
bone
l
l
,
R
.
S
a
l
a
khut
di
nov,
a
nd
Q
.
V
.
L
e
,
“
X
L
N
e
t
:
ge
ne
r
a
l
i
z
e
d
a
ut
or
e
gr
e
s
s
i
ve
pr
e
t
r
a
i
ni
ng
f
or
l
a
ngua
ge
unde
r
s
t
a
ndi
ng,”
A
dv
anc
e
s
i
n N
e
u
r
al
I
nf
or
m
at
i
on P
r
oc
e
s
s
i
ng Sy
s
t
e
m
s
,
vol
. 32, 2019.
[
28]
C
.
R
a
f
f
e
l
e
t
al
.
,
“
E
xpl
or
i
ng
t
he
l
i
m
i
t
s
of
t
r
a
ns
f
e
r
l
e
a
r
ni
ng
w
i
t
h
a
uni
f
i
e
d
t
e
xt
-
to
-
t
e
xt
t
r
a
ns
f
or
m
e
r
,”
J
our
nal
of
M
ac
hi
ne
L
e
a
r
ni
ng
R
e
s
e
ar
c
h
, vol
. 21, 2020.
[
29]
A
. M
. R
us
h, S
.
C
hopr
a
, a
nd J
. W
e
s
t
on, “
A
ne
ur
a
l
a
t
t
e
nt
i
on m
ode
l
f
or
a
bs
t
r
a
c
t
i
v
e
s
e
nt
e
nc
e
s
um
m
a
r
i
z
a
t
i
on,”
ar
X
i
v
, S
e
p. 2015.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
2
,
A
pr
il
20
2
5
:
1587
-
1596
1596
[
30]
A
. C
e
l
i
kyi
l
m
a
z
, A
.
B
o
s
s
e
l
ut
,
X
. H
e
,
a
nd Y
.
C
hoi
,
“
D
e
e
p
c
om
m
uni
c
a
t
i
ng a
g
e
nt
s
f
or
a
bs
t
r
a
c
t
i
ve
s
um
m
a
r
i
z
a
t
i
on,”
i
n
P
r
oc
e
e
di
ng
s
o
f
t
he
2018
C
onf
e
r
e
nc
e
of
t
he
N
o
r
t
h
A
m
e
r
i
c
an
C
hapt
e
r
of
t
he
A
s
s
oc
i
at
i
on
f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
:
H
um
an
L
anguage
T
e
c
hnol
ogi
e
s
, 2018, pp. 1662
–
1675
, doi
:
10.18653/
v1/
n18
-
1150.
[
31]
S
.
S
ukhba
a
t
a
r
,
A
.
S
z
l
a
m
,
a
nd
R
.
F
e
r
gus
,
“
L
e
a
r
ni
ng
m
ul
t
i
a
ge
nt
c
om
m
uni
c
a
t
i
on
w
i
t
h
ba
c
kpr
opa
ga
t
i
on,”
A
dv
anc
e
s
i
n
N
e
u
r
al
I
nf
or
m
at
i
on P
r
oc
e
s
s
i
ng Sy
s
t
e
m
s
, pp. 2252
–
2260, 2016.
[
32]
R
. P
a
ul
us
,
C
. X
i
ong, a
nd
R
. S
oc
he
r
, “
A
de
e
p r
e
i
nf
or
c
e
d m
ode
l
f
or
a
bs
t
r
a
c
t
i
ve
s
um
m
a
r
i
z
a
t
i
on,”
i
n
6t
h I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
o
n
L
e
ar
ni
ng R
e
pr
e
s
e
nt
at
i
on
s
, I
C
L
R
2018
-
C
onf
e
r
e
nc
e
T
r
ac
k
P
r
oc
e
e
di
ng
s
, 2018,
pp. 1
–
13.
[
33]
S
. N
a
r
a
ya
n, S
. B
. C
ohe
n, a
nd
M
. L
a
pa
t
a
, “
D
on’
t
gi
ve
m
e
t
he
de
t
a
i
l
s
, j
us
t
t
he
s
u
m
m
a
r
y!
t
opi
c
-
a
w
a
r
e
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
ks
f
or
e
xt
r
e
m
e
s
um
m
a
r
i
z
a
t
i
on,”
i
n
P
r
o
c
e
e
di
ngs
of
t
he
2018
C
onf
e
r
e
nc
e
on
E
m
pi
r
i
c
al
M
e
t
hods
i
n
N
at
ur
al
L
anguage
P
r
oc
e
s
s
i
ng
,
E
M
N
L
P
2018
, 2018, pp. 1797
–
1807
, doi
:
10.18653/
v1
/
d18
-
1206.
[
34]
Y
. L
i
u a
nd
M
.
L
a
p
a
t
a
, “
T
e
xt
s
um
m
a
r
i
z
a
t
i
on w
i
t
h pr
e
t
r
a
i
ne
d e
n
c
ode
r
s
,”
i
n
E
M
N
L
P
-
I
J
C
N
L
P
2019
-
2019 C
onf
e
r
e
nc
e
on E
m
pi
r
i
c
a
l
M
e
t
hods
i
n
N
at
ur
al
L
anguage
P
r
o
c
e
s
s
i
ng
and
9t
h
I
nt
e
r
nat
i
onal
J
oi
nt
C
onf
e
r
e
nc
e
on
N
at
ur
al
L
anguage
P
r
oc
e
s
s
i
ng,
P
r
oc
e
e
di
ng
s
of
t
he
C
onf
e
r
e
nc
e
,
A
ug. 2019, pp. 3730
–
3740
, doi
:
10.18653/
v1/
d19
-
1387.
[
35]
K
.
M
.
H
e
r
m
a
nn
e
t
al
.
,
“
T
e
a
c
hi
ng
m
a
c
hi
ne
s
t
o
r
e
a
d
a
nd
c
om
pr
e
he
nd,”
i
n
A
dv
anc
e
s
i
n
N
e
u
r
al
I
nf
or
m
at
i
on
P
r
oc
e
s
s
i
ng
S
y
s
t
e
m
s
,
C
a
m
br
i
dge
:
M
I
T
P
r
e
s
s
,
2015, pp. 1693
–
1701.
[
36]
E
.
S
a
ndha
us
,
“
T
he
ne
w
yor
k
t
i
m
e
s
a
nnot
a
t
e
d
c
or
pus
,
”
A
bac
us
D
a
t
a
N
e
t
w
or
k
,
V
1,
2008
.
[
O
nl
i
ne
]
.
A
va
i
l
a
bl
e
:
ht
t
ps
:
/
/
hdl
.ha
ndl
e
.ne
t
/
11272.1/
A
B
2/
G
Z
C
6P
L
[
37]
W
.
Q
i
e
t
al
.
,
“
P
r
ophe
t
N
e
t
:
pr
e
di
c
t
i
ng
f
ut
u
r
e
n
-
gr
a
m
f
or
s
e
que
nc
e
-
to
-
s
e
que
nc
e
pr
e
-
t
r
a
i
ni
ng,”
i
n
F
i
ndi
ngs
of
t
he
A
s
s
oc
i
at
i
on
f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
F
i
ndi
ngs
of
A
C
L
:
E
M
N
L
P
2020
, 2020, pp. 2401
–
24
10, doi
:
10.18653/
v1/
2020.f
i
ndi
ngs
-
e
m
nl
p.217.
[
38]
Z
.
A
l
t
a
n,
“
A
T
ur
ki
s
h
a
ut
om
a
t
i
c
t
e
xt
s
um
m
a
r
i
z
a
t
i
on
s
y
s
t
e
m
,”
i
n
P
r
oc
e
e
di
ng
s
of
t
he
I
A
ST
E
D
I
nt
e
r
nat
i
onal
C
onf
e
r
e
n
c
e
:
A
ppl
i
e
d
I
nf
or
m
at
i
c
s
, 2004, pp. 311
–
316.
[
39]
M
.
K
ut
l
u,
C
.
C
i
ǧi
r
,
a
nd
I
.
C
i
c
e
kl
i
,
“
G
e
ne
r
i
c
t
e
xt
s
um
m
a
r
i
z
a
t
i
on
f
or
t
u
r
ki
s
h,”
C
om
put
e
r
J
our
nal
,
vol
.
53,
no.
8,
pp.
1315
–
1323,
2010, doi
:
10.1093/
c
om
j
nl
/
bxp124.
[
40]
Y
.
S
.
K
a
r
t
a
l
a
nd
M
.
K
ut
l
u,
“
M
a
c
hi
ne
l
e
a
r
ni
ng
ba
s
e
d
t
e
xt
s
um
m
a
r
i
z
a
t
i
on
f
or
t
ur
ki
s
h
ne
w
s
,”
i
n
2020
28t
h
Si
gnal
P
r
oc
e
s
s
i
ng
and
C
om
m
uni
c
at
i
ons
A
ppl
i
c
at
i
ons
C
onf
e
r
e
nc
e
, SI
U
2020
-
P
r
oc
e
e
di
ng
s
, I
E
E
E
, 202
0, pp.
1
–
4, doi
:
10.1109/
S
I
U
49456.2020.9302096.
[
41]
M
.
O
z
s
oy,
I
.
C
i
c
e
kl
i
,
a
nd
F
.
A
l
pa
s
l
a
n,
“
T
e
xt
s
um
m
a
r
i
z
a
t
i
on
of
t
ur
ki
s
h
t
e
xt
s
us
i
ng
l
a
t
e
nt
s
e
m
a
nt
i
c
a
na
l
ys
i
s
,”
i
n
P
r
oc
e
e
di
ng
s
of
t
he
23r
d I
nt
e
r
nat
i
onal
C
onf
e
r
e
n
c
e
on C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
(
C
ol
i
ng 2010)
, 20
10, pp. 869
–
876.
[
42]
F
.
C
.
P
e
m
be
,
“
A
ut
om
a
t
e
d
que
r
y
-
bi
a
s
e
d
a
nd
s
t
r
uc
t
ur
e
-
pr
e
s
e
r
vi
ng
doc
um
e
nt
s
um
m
a
r
i
z
a
t
i
on
f
or
w
e
b
s
e
a
r
c
h
t
a
s
ks
,”
P
h.D
T
he
s
i
s
,
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
E
ngi
ne
e
r
i
ng,
B
oğa
z
i
ç
i
U
ni
ve
r
s
i
t
y, I
s
t
a
nbul
, T
ur
ke
y, 2
010.
[
43]
A
.
G
ür
a
n,
“
A
ut
om
a
t
i
c
t
e
x
s
um
m
a
r
i
z
a
t
i
on
s
ys
t
e
m
,”
P
h.D
T
he
s
i
s
,
D
e
pa
r
t
m
e
nt
o
f
C
om
put
e
r
,
Y
ı
l
d
ı
z
T
e
c
hni
c
a
l
U
ni
ve
r
s
i
t
y,
I
s
t
a
nbul
,
T
ur
ke
y, 2013.
[
44]
E
.
K
a
r
a
koc
a
nd
B
.
Y
i
l
m
a
z
,
“
D
e
e
p
l
e
a
r
ni
ng
ba
s
e
d
a
b
s
t
r
a
c
t
i
ve
t
ur
ki
s
h
ne
w
s
s
um
m
a
r
i
z
a
t
i
on,”
i
n
27t
h
Si
gnal
P
r
oc
e
s
s
i
ng
an
d
C
om
m
uni
c
at
i
ons
A
ppl
i
c
at
i
ons
C
onf
e
r
e
nc
e
, SI
U
2019
, I
E
E
E
, 2019, pp. 1
–
4
, doi
:
10.1109/
S
I
U
.2019.8806510.
[
45]
C.
-
Y
. L
i
n, “
R
O
U
G
E
:
a
pa
c
k
a
ge
f
or
a
ut
om
a
t
i
c
e
va
l
ua
t
i
on of
s
um
m
a
r
i
e
s
,”
i
n
T
e
x
t
Sum
m
ar
i
z
at
i
on B
r
anc
he
s
O
ut
, 2004, pp. 74
–
81.
[
46]
V
.
A
hui
r
,
L
.
F
.
H
ur
t
a
do,
J
.
Á
.
G
on
z
á
l
e
z
,
a
nd
E
.
S
e
ga
r
r
a
,
“
N
a
s
c
a
a
nd
n
a
s
e
s
:
t
w
o
m
onol
i
ngua
l
pr
e
-
t
r
a
i
ne
d
m
ode
l
s
f
or
a
b
s
t
r
a
c
t
i
ve
s
um
m
a
r
i
z
a
t
i
on i
n c
a
t
a
l
a
n a
nd s
p
a
ni
s
h,”
A
ppl
i
e
d Sc
i
e
nc
e
s
, vol
. 11, no. 21, 2021,
doi
:
10.3390/
a
pp11219872.
[
47]
M
. P
a
nt
a
nd A
.
C
hopr
a
, “
M
ul
t
i
l
i
ngua
l
f
i
na
nc
i
a
l
doc
um
e
nt
a
t
i
on s
um
m
a
r
i
z
a
t
i
on
by t
e
a
m
_t
r
e
de
nc
e
f
or
F
N
S
2022
,”
i
n
P
r
oc
e
e
di
ngs
of
t
he
4t
h F
i
nanc
i
al
N
ar
r
at
i
v
e
P
r
oc
e
s
s
i
ng W
or
k
s
hop
, 2022, pp. 112
–
115.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
S
Neda
Alipour
holds
a
doctor
of
management
information
system
s
degree
fro
m
Ata
türk
University,
Türkiye
in
2022.
He
also
received
his
B.Sc.
(
information
technology
engineerin
g
)
from
Tabriz
University,
Iran
in
2011
and
M.Sc.
(MIS)
f
rom
Atatürk
University,
Türkiye
in
2017
and
respectively.
His
research
includes
natural
la
nguage
rocessing,
deep
learning,
e
-
commerce,
and
e
-
government.
Sh
e
can
be
contacted
at
email:
nedaalipoor
@
yahoo.com or neda.alipour14@ogr.atauni.edu.tr.
Serdar
Aydın
is
currently
an
associate
professors
hip
in
Departme
nt
of
Softwa
re
Engineering
in
Türkiye,
Atatürk
University.
His
research
includ
es
social
sciences
and
humanities,
science,
technology,
and
soci
ety.
He
has
published
over
70
papers
in
international
journals an
d confer
ences.
He can be contacted at email:
serdar@atauni.edu.tr.
Evaluation Warning : The document was created with Spire.PDF for Python.