I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
,
p
p
.
4
9
6
5
~
4
9
7
5
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
13
.i
4
.
p
p
4
9
6
5
-
4
9
7
5
4965
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
M
a
la
y
ph
o
neme
-
ba
sed subwo
rd n
e
ws hea
dline gen
er
a
tor for
lo
w
-
reso
urce lang
ua
g
e
Yeo
ng
T
s
a
nn
P
h
ua
1,
2
,
K
wa
ng
H
o
o
i
Yew
1
,
M
o
hd
F
a
dzil H
a
s
s
a
n
1
,
M
a
t
t
hew
T
eo
w
Yo
k
Wo
o
i
3
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
a
n
d
I
n
f
o
r
m
a
t
i
o
n
S
c
i
e
n
c
e
,
U
n
i
v
e
r
s
i
t
i
T
e
k
n
o
l
o
g
i
P
ETR
O
N
A
S
,
S
e
r
i
I
sk
a
n
d
a
r
,
M
a
l
a
y
si
a
2
S
c
h
o
o
l
o
f
D
i
p
l
o
m
a
S
t
u
d
i
e
s,
Ta
y
l
o
r
’
s
C
o
l
l
e
g
e
,
S
u
b
a
n
g
Ja
y
a
,
M
a
l
a
y
si
a
3
U
n
i
v
e
r
s
i
t
y
P
a
r
t
n
e
r
sh
i
p
(
C
o
v
e
n
t
r
y
U
n
i
v
e
r
si
t
y
)
,
P
S
B
A
c
a
d
e
m
y
,
R
a
f
f
l
e
s
,
S
i
n
g
a
p
o
r
e
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Dec
18
,
2
0
2
3
R
ev
is
ed
May
1
7
,
2
0
2
4
Acc
ep
ted
J
u
n
1
,
2
0
2
4
Th
e
b
o
o
m
i
n
g
o
f
tec
h
n
o
lo
g
y
h
a
s
sig
n
ifi
c
a
n
t
ly
in
c
re
a
se
d
th
e
a
m
o
u
n
t
o
f
n
e
ws
a
rti
c
les
fo
r
re
a
d
e
rs.
Th
e
h
e
a
d
li
n
e
o
f
n
e
ws
p
lay
s
a
n
e
ss
e
n
ti
a
l
r
o
le
i
n
a
tt
ra
c
ti
n
g
re
a
d
e
rs.
Trad
it
io
n
a
ll
y
,
c
ra
fti
n
g
th
e
n
e
ws
h
e
a
d
li
n
e
is
a
m
a
n
u
a
l
tas
k
a
t
th
e
n
e
ws
d
e
sk
.
Th
e
m
o
ti
v
a
ti
o
n
o
f
th
is
p
a
p
e
r
is
to
a
d
d
re
ss
th
e
issu
e
s
fa
c
e
d
in
lo
w
-
re
so
u
rc
e
lan
g
u
a
g
e
s,
su
c
h
a
s
th
e
M
a
lay
lan
g
u
a
g
e
.
T
h
e
m
a
in
c
o
n
tri
b
u
t
io
n
o
f
t
h
is
p
a
p
e
r
is
a
n
e
w
h
y
b
r
id
m
o
d
e
l
b
a
s
e
d
o
n
e
x
trac
ti
v
e
-
a
n
d
a
b
stra
c
ti
v
e
-
b
a
se
d
tex
t
su
m
m
a
riza
ti
o
n
with
t
h
e
in
teg
ra
ti
o
n
o
f
a
g
e
o
g
ra
p
h
ica
l
li
n
g
u
isti
c
s
m
o
d
e
l;
a
M
a
lay
p
h
o
n
e
m
e
-
b
a
se
d
su
b
wo
r
d
e
m
b
e
d
d
i
n
g
h
a
s
b
e
e
n
d
e
v
e
lo
p
e
d
t
o
so
lv
e
t
h
e
c
o
m
p
lex
m
o
r
p
h
o
l
o
g
ica
l
issu
e
i
n
th
e
M
a
lay
lan
g
u
a
g
e
-
b
a
se
d
c
o
m
p
u
tati
o
n
a
l
li
n
g
u
isti
c
a
p
p
li
c
a
ti
o
n
s.
T
h
e
e
x
p
e
rime
n
t
i
n
v
o
lv
e
s
v
a
rio
u
s
se
q
u
e
n
c
e
-
to
-
se
q
u
e
n
c
e
(
s
e
q
2
se
q
)
m
o
d
e
ls
to
g
e
n
e
ra
te
th
e
M
a
lay
n
e
ws
h
e
a
d
li
n
e
s
.
Be
sid
e
s
th
a
t,
t
h
e
o
u
t
-
of
-
v
o
c
a
b
u
lary
(OO
V)
is
a
ss
e
ss
e
d
in
th
e
m
o
d
e
ls.
F
ro
m
th
e
e
x
p
e
rime
n
t,
t
h
e
p
ro
p
o
se
d
h
y
b
ri
d
t
e
x
t
su
m
m
a
riza
ti
o
n
m
o
d
e
l
s
h
o
ws
s
ig
n
ifi
c
a
n
t
imp
ro
v
e
m
e
n
t
o
v
e
r
th
e
b
a
se
li
n
e
m
o
d
e
ls
a
b
o
v
e
1
1
.
0
0
in
ROU
G
E
-
1
,
4
.
0
0
ROU
G
E
-
2
,
a
n
d
1
1
.
0
0
in
ROU
G
E
-
L.
T
h
e
p
ro
p
o
se
d
m
o
d
e
l
c
a
n
re
d
u
c
e
th
e
OO
V
ra
te
to
b
e
l
o
w 1
5
%
.
K
ey
w
o
r
d
s
:
L
o
w
-
r
eso
u
r
ce
la
n
g
u
a
g
e
Ma
lay
lan
g
u
ag
e
Natu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
New
s
h
ea
d
lin
e
g
en
er
ato
r
Su
b
wo
r
d
em
b
ed
d
i
n
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Yeo
n
g
T
s
an
n
Ph
u
a
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
a
n
d
I
n
f
o
r
m
atio
n
Scien
ce
,
Un
i
v
er
s
iti T
ek
n
o
lo
g
i PE
T
R
ONAS
Ser
i I
s
k
an
d
ar
,
Per
a
k
Dar
u
l Ri
d
zu
an
3
2
6
1
0
,
Ma
lay
s
ia
E
m
ail: y
eo
n
g
_
1
7
0
0
8
2
5
6
@
u
tp
.
ed
u
.
m
y
1.
I
NT
RO
D
UCT
I
O
N
I
n
th
e
wo
r
l
d
o
f
j
o
u
r
n
alis
m
,
n
e
ws h
ea
d
lin
es ser
v
e
as th
e
ag
en
ts
to
p
r
o
v
id
e
th
e
ess
en
ce
o
f
n
e
ws
ar
ticles
[
1
]
.
T
h
e
h
ea
d
lin
e
p
la
y
s
a
p
iv
o
tal
r
o
le
in
p
r
o
v
id
in
g
th
e
m
a
in
p
o
i
n
t
o
f
t
h
e
n
ews
s
to
r
y
wh
ile
attr
ac
tin
g
an
d
m
o
tiv
atin
g
r
ea
d
er
s
t
o
ex
p
lo
r
e
th
e
n
ews
co
n
ten
t
f
u
r
t
h
er
.
M
o
r
eo
v
e
r
,
a
well
-
cr
af
ted
h
ea
d
li
n
e
will
r
ed
u
ce
th
e
r
ea
d
er
’
s
co
g
n
itiv
e
b
u
r
d
en
wh
ile
r
ea
d
in
g
th
e
n
ews
[
2
]
.
T
h
e
r
ef
o
r
e,
co
n
s
tr
u
ctin
g
ef
f
icie
n
t
an
d
ac
cu
r
ate
n
ews
h
ea
d
lin
es
is
a
cr
u
cial
jo
u
r
n
alis
tic
s
k
ill
b
ec
au
s
e
it
s
h
ap
es
h
o
w
th
e
n
ews
is
co
n
s
u
m
ed
,
s
h
ar
ed
,
an
d
p
e
r
ce
iv
ed
b
y
th
e
au
d
ien
ce
.
Ho
we
v
er
,
in
th
e
cu
r
r
en
t
f
ast
-
p
ac
e
d
n
ews
en
v
ir
o
n
m
en
t,
jo
u
r
n
al
is
ts
an
d
ed
it
o
r
s
o
f
ten
f
ac
e
tim
e
co
n
s
tr
ain
ts
.
Un
d
er
th
ese
cir
cu
m
s
tan
ce
s
,
au
to
m
atic
n
ews
h
ea
d
lin
e
g
en
er
atio
n
(
NHG)
s
y
s
te
m
s
h
elp
to
s
tr
ea
m
lin
e
th
e
p
r
o
ce
s
s
an
d
s
av
e
tim
e
as
w
ell
as
r
e
s
o
u
r
ce
s
b
y
q
u
ick
ly
p
r
o
d
u
cin
g
h
ea
d
lin
es
b
ased
o
n
th
e
co
n
ten
t
o
f
th
e
n
ews
s
to
r
y.
T
h
e
NHG
is
co
n
s
id
er
ed
a
ty
p
e
o
f
tex
t
s
u
m
m
ar
izatio
n
.
T
ex
t
s
u
m
m
ar
izatio
n
ca
n
o
cc
u
r
f
o
r
b
o
th
s
in
g
le
o
r
m
u
ltip
le
d
o
cu
m
en
ts
[
3
]
.
Gen
er
ally
,
th
er
e
ar
e
two
a
p
p
r
o
ac
h
es
ad
o
p
ted
in
NHG,
th
e
ex
tr
ac
tiv
e
an
d
th
e
ab
s
tr
ac
tiv
e
ap
p
r
o
ac
h
es [
4
]
.
B
o
th
ap
p
r
o
ac
h
es a
im
to
p
r
o
d
u
c
e
s
h
o
r
t y
et
ac
cu
r
ate
h
ea
d
lin
es th
at
co
n
v
ey
th
e
m
ain
id
ea
o
f
th
e
n
ews
co
n
ten
t.
T
h
e
ex
tr
ac
tiv
e
ap
p
r
o
ac
h
aim
s
at
s
ev
er
al
im
p
o
r
tan
t
s
en
ten
ce
s
an
d
ap
p
lies
s
en
ten
ce
co
m
p
r
ess
io
n
tech
n
iq
u
es
to
p
r
o
d
u
ce
th
e
h
ea
d
lin
e
[
4
]
.
Du
e
to
th
e
r
eu
s
e
o
f
co
n
ten
t
wo
r
d
s
in
th
e
h
ea
d
lin
e,
d
is
to
r
tio
n
o
f
m
ea
n
in
g
will
o
c
cu
r
in
th
e
h
ea
d
lin
e
p
r
o
d
u
ce
d
[
5
]
.
T
h
e
a
b
s
tr
ac
tiv
e
ap
p
r
o
ac
h
ap
p
lies
ad
v
an
ce
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
9
6
5
-
4
9
7
5
4966
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
(
NL
P)
tech
n
iq
u
es
to
p
r
o
d
u
ce
th
e
h
ea
d
lin
e.
So
m
e
o
f
th
es
e
tech
n
iq
u
es
allo
w
p
ar
ap
h
r
asin
g
,
s
y
n
o
n
y
m
o
u
s
s
u
b
s
titu
tio
n
s
,
an
d
s
en
ten
ce
c
o
n
t
r
ac
tio
n
s
[
6
]
.
T
h
e
n
eu
r
al
n
etw
o
r
k
-
b
ased
atten
tio
n
m
ec
h
an
is
m
en
co
d
er
-
d
ec
o
d
er
m
o
d
els [
7
]
h
av
e
d
em
o
n
s
t
r
ated
im
p
r
ess
iv
e
r
esu
lts
s
u
ch
as
in
[
8
]
,
[
9
]
.
As
NHG
b
ec
o
m
es
a
n
im
p
o
r
ta
n
t
task
in
NL
P
[
6
]
,
r
esear
ch
o
n
an
d
d
ev
elo
p
m
en
t o
f
th
e
s
y
s
tem
s
m
ain
ly
f
o
cu
s
ed
o
n
co
m
m
o
n
l
y
s
p
o
k
en
lan
g
u
a
g
es
s
u
ch
as
E
n
g
lis
h
,
w
h
ile
o
v
er
lo
o
k
in
g
lo
w
r
eso
u
r
ce
lan
g
u
a
g
es
(
L
R
L
)
.
Mo
s
t
o
f
th
e
s
tate
-
of
-
th
e
-
ar
t
a
p
p
r
o
ac
h
es
in
NHG
wer
e
ab
le
to
p
r
o
d
u
ce
im
p
r
ess
iv
e
r
esu
lts
with
tr
ain
i
n
g
u
s
in
g
lar
g
e
d
atasets
s
u
ch
as
Gig
awo
r
d
,
wh
ich
is
in
E
n
g
lis
h
[
1
0
]
,
[
1
1
]
.
I
n
co
n
tr
ast,
L
R
L
,
lik
e
m
a
n
y
Asi
an
lan
g
u
ag
es
in
clu
d
in
g
th
e
Ma
lay
lan
g
u
ag
e,
ar
e
o
f
ten
o
v
er
lo
o
k
ed
b
ec
a
u
s
e
th
ey
lack
lin
g
u
is
tic
r
eso
u
r
ce
s
s
u
ch
as
p
ar
t
o
f
s
p
ee
ch
(
POS)
tag
g
er
s
an
d
co
r
p
o
r
a
[
1
2
]
T
h
e
lim
ited
av
ailab
ilit
y
o
f
L
R
L
d
ataset
h
as
led
t
o
ch
allen
g
es
in
th
e
d
ev
elo
p
m
e
n
t
o
f
au
to
m
atic
h
ea
d
lin
e
g
en
er
atio
n
s
p
ec
if
ical
ly
tailo
r
ed
f
o
r
L
R
L
.
T
h
is
s
itu
atio
n
is
f
u
r
th
er
co
m
p
licated
b
y
th
e
Ma
lay
lan
g
u
ag
e
’
s
r
ich
m
o
r
p
h
o
lo
g
ical
s
tr
u
ctu
r
e,
wh
ich
o
f
ten
r
esu
lts
in
a
h
ig
h
o
u
t
-
of
-
v
o
ca
b
u
lar
y
(
OOV)
r
ate
wh
en
t
r
ad
itio
n
al
wo
r
d
e
m
b
ed
d
in
g
m
e
th
o
d
s
ar
e
em
p
lo
y
ed
.
T
o
n
av
ig
ate
th
ese
o
b
s
tacle
s
,
th
is
s
tu
d
y
in
tr
o
d
u
ce
s
a
Ma
lay
s
alien
t
s
en
ten
ce
ex
tr
ac
to
r
th
at
lev
er
ag
es
s
en
ten
ce
e
x
tr
ac
t
io
n
m
eth
o
d
s
to
d
is
till
ess
en
tial c
o
n
ten
t,
th
er
eb
y
s
tr
ea
m
lin
in
g
b
o
th
th
e
m
o
d
el
tr
ain
i
n
g
p
r
o
ce
s
s
an
d
th
e
h
ea
d
lin
e
g
en
er
atio
n
task
.
I
n
r
esp
o
n
s
e
to
th
e
lim
itatio
n
s
im
p
o
s
ed
b
y
th
e
a
v
ailab
ilit
y
o
f
r
eso
u
r
ce
s
an
d
th
e
in
tr
in
s
ic
lan
g
u
a
g
e
f
ea
tu
r
es,
th
e
au
th
o
r
s
p
r
o
p
o
s
e
d
a
s
eq
u
en
ce
-
to
-
s
eq
u
en
ce
(
s
e
q
2
s
eq
)
d
ee
p
lear
n
in
g
-
b
ased
m
o
d
el
n
am
ed
Ma
lay
s
alien
t
ab
s
tr
ac
tiv
e
s
u
m
m
ar
izer
(
MSAS)
.
I
t
is
b
eliev
ed
th
at
in
n
o
v
ativ
e
a
p
p
r
o
ac
h
es
to
N
HG
f
o
r
th
e
Ma
lay
lan
g
u
ag
e
ca
n
o
f
f
er
p
r
o
m
is
in
g
s
o
lu
tio
n
s
.
T
h
is
r
esear
ch
ad
o
p
ts
a
h
y
b
r
id
s
eq
2
s
eq
ap
p
r
o
ac
h
o
f
ex
tr
ac
tiv
e
an
d
ab
s
tr
ac
tiv
e
m
eth
o
d
s
tailo
r
e
d
t
o
th
e
Ma
lay
lin
g
u
is
tic
co
n
tex
t
.
T
h
e
Ma
lay
p
h
o
n
em
e
-
b
ased
s
u
b
wo
r
d
em
b
ed
d
in
g
h
as
b
ee
n
em
p
lo
y
e
d
to
ef
f
ec
tiv
ely
ad
d
r
ess
th
e
p
er
v
asiv
e
OO
V
is
s
u
e
[
1
3
]
,
s
u
b
s
tan
tially
r
ed
u
cin
g
th
e
em
b
e
d
d
in
g
lay
er
’
s
v
o
ca
b
u
lar
y
s
ize
with
o
u
t
co
m
p
r
o
m
is
in
g
t
h
e
m
o
d
el'
s
lin
g
u
is
tic
co
m
p
r
eh
e
n
s
io
n
[
1
4
]
.
T
h
r
o
u
g
h
th
ese
m
eth
o
d
o
l
o
g
ical
ad
v
an
ce
m
en
ts
,
th
is
s
tu
d
y
n
o
t
o
n
ly
co
n
tr
ib
u
te
s
to
th
e
b
o
d
y
o
f
k
n
o
wled
g
e
in
NL
P
b
u
t
also
o
f
f
er
s
p
r
ac
tical
s
o
lu
tio
n
s
f
o
r
a
u
to
m
at
ed
h
ea
d
lin
e
g
en
er
ati
o
n
in
th
e
c
o
n
tex
t
o
f
L
R
L
.
T
h
e
ex
p
er
im
en
t
is
co
n
d
u
cted
u
s
in
g
a
d
ataset
o
f
4
5
,
0
0
0
Ma
lay
lan
g
u
ag
e
n
ews sto
r
ies an
d
th
eir
h
ea
d
lin
es.
T
h
e
r
el
ate
d
wo
r
k
is
d
esc
r
i
b
e
d
i
n
s
e
cti
o
n
2
.
Se
cti
o
n
3
p
r
o
v
i
d
es
th
e
p
r
o
p
o
s
e
d
MS
AS
m
o
d
e
l
a
r
ch
i
te
ct
u
r
e.
T
h
e
e
x
p
e
r
i
m
e
n
t
s
et
u
p
,
ex
ec
u
t
i
o
n
,
an
d
r
es
u
l
t
a
r
e
d
is
c
u
s
s
e
d
i
n
s
ec
ti
o
n
4
.
T
h
e
c
o
n
cl
u
s
i
o
n
is
p
r
ese
n
te
d
i
n
s
ec
t
io
n
5.
2.
RE
L
AT
E
D
WO
RK
A
g
o
o
d
n
ews
h
ea
d
lin
e
is
ab
le
to
co
n
v
ey
a
co
n
cise
s
u
m
m
a
r
y
o
f
th
e
n
ews
ar
ticle
[
1
5
]
,
[
1
6
]
a
n
d
attr
ac
t
r
ea
d
er
s
’
atten
tio
n
.
T
h
e
p
r
o
ce
s
s
o
f
au
to
m
atic
h
ea
d
lin
e
g
e
n
er
ati
o
n
h
as
b
ee
n
a
r
esear
ch
f
o
cu
s
in
NL
P.
T
h
is
s
ec
tio
n
will p
r
esen
t th
e
p
r
ev
io
u
s
wo
r
k
o
n
NHG
an
d
t
h
e
Ma
lay
lan
g
u
ag
e
NHG.
2
.
1
.
New
s
hea
dli
ne
g
ener
a
t
io
n
I
n
t
h
e
e
ar
ly
y
e
ar
s
,
Gatt
a
n
i
[
1
7
]
c
ate
g
o
r
iz
ed
NH
G
i
n
t
o
th
r
e
e
b
r
o
a
d
ap
p
r
o
ac
h
es:
r
u
le
d
-
b
ase
d
,
s
tatis
t
ics
-
b
ase
d
,
a
n
d
s
u
m
m
ar
i
z
ati
o
n
-
b
as
ed
.
T
h
e
r
u
le
-
b
as
ed
a
p
p
r
o
ac
h
d
et
ec
ts
a
n
d
co
m
p
r
ess
es
im
p
o
r
ta
n
t
p
a
r
ts
o
f
an
ar
t
icl
e
u
s
i
n
g
h
a
n
d
c
r
a
f
t
e
d
li
n
g
u
is
tic
al
r
u
les
s
u
c
h
as
h
ed
g
e
tr
i
m
m
er
[
1
6
]
.
E
v
e
n
t
h
o
u
g
h
t
h
i
s
a
p
p
r
o
ac
h
is
s
i
m
p
le
y
et
l
ig
h
t
wei
g
h
t
,
i
t
is
n
o
t
ab
le
t
o
d
is
c
o
v
e
r
t
h
e
c
o
m
p
l
e
x
r
el
ati
o
n
s
h
i
p
s
in
t
h
e
t
ex
t.
He
n
ce
,
t
h
e
m
es
s
ag
e
i
n
t
h
e
h
ea
d
l
in
e
d
e
v
i
ates
f
r
o
m
th
e
o
r
i
g
i
n
al
c
o
n
ten
t
[
1
8
]
.
Me
a
n
w
h
ile
,
th
e
s
t
ati
s
tica
l
-
b
as
e
d
a
p
p
r
o
ac
h
u
s
es
s
ta
tis
tic
al
m
e
th
o
d
s
t
o
d
is
c
o
v
e
r
th
e
c
o
r
r
e
lat
io
n
b
etw
ee
n
t
h
e
w
o
r
d
s
i
n
th
e
h
ea
d
li
n
es
a
n
d
i
n
t
h
e
n
ews
c
o
n
te
n
t
.
S
o
m
e
o
f
th
e
e
ar
ly
a
d
o
p
ti
o
n
s
o
f
t
h
is
a
p
p
r
o
ac
h
ar
e
n
aï
v
e
B
a
y
e
s
[
1
0
]
an
d
u
n
s
u
p
er
v
is
e
d
t
o
p
ic
d
is
co
v
er
y
(
U
T
D
)
b
y
Do
r
r
e
t
a
l.
[
1
9
]
.
T
h
e
wea
k
n
ess
o
f
th
ese
m
o
d
els
is
th
e
lar
g
e
tr
ai
n
in
g
d
ataset
r
eq
u
ir
em
e
n
t.
On
th
e
o
th
er
h
an
d
,
th
e
s
u
m
m
ar
izati
o
n
-
b
ased
ap
p
r
o
ac
h
r
eg
ar
d
s
h
ea
d
lin
es
as
v
er
y
s
h
o
r
t
s
u
m
m
ar
ies
o
f
c
o
n
ten
t.
T
r
ad
iti
o
n
al
tex
t
s
u
m
m
a
r
izatio
n
a
p
p
r
o
ac
h
es
wer
e
ad
o
p
ted
to
p
r
o
d
u
ce
a
o
n
e
-
lin
e
s
en
ten
c
e
th
at
ca
n
b
e
r
e
g
ar
d
e
d
as
a
h
e
ad
lin
e
[
2
0
]
.
T
h
is
ap
p
r
o
ac
h
r
eq
u
ir
es
m
u
ltip
le
s
tep
s
o
r
co
m
b
in
ed
s
tep
s
o
f
s
en
ten
ce
s
elec
tio
n
an
d
co
m
p
r
ess
io
n
.
T
h
e
wea
k
n
ess
o
f
th
e
ap
p
r
o
ac
h
is
th
at
th
e
co
m
p
r
ess
io
n
ap
p
r
o
ac
h
is
n
o
t
s
u
itab
le
f
o
r
p
r
o
d
u
cin
g
h
ea
d
li
n
es
th
at
ar
e
o
n
ly
1
0
%
o
r
less
th
an
th
e
o
r
ig
in
al
n
ews
ar
ticles.
B
esid
es th
at,
th
e
o
u
tp
u
t m
ay
f
ac
e
co
n
tex
tu
al
d
is
to
r
tio
n
d
u
e
t
o
th
e
r
eu
s
ed
wo
r
d
s
.
L
ater
,
th
e
NHG
h
as
b
ee
n
r
e
-
ca
teg
o
r
ized
in
to
e
x
tr
ac
tiv
e
an
d
ab
s
tr
ac
tiv
e
ap
p
r
o
ac
h
es
[
3
]
.
B
o
th
th
e
r
u
le
-
b
ased
an
d
s
tatis
tical
-
b
ase
d
f
all
in
to
th
e
ex
tr
ac
tiv
e
ap
p
r
o
ac
h
.
T
h
e
ab
s
tr
ac
tiv
e
ap
p
r
o
ac
h
,
h
o
wev
er
,
is
ab
le
to
m
ain
tain
an
d
f
o
cu
s
o
n
th
e
m
a
in
id
ea
o
f
th
e
n
ews
wh
en
g
en
er
atin
g
th
e
h
ea
d
lin
e
[
2
1
]
.
T
h
e
g
en
er
ated
h
ea
d
lin
e
m
ay
n
o
t
co
n
tain
wo
r
d
s
o
r
r
e
d
u
n
d
an
t
wo
r
d
s
f
r
o
m
th
e
o
r
ig
in
al
n
ews
ar
ticle,
wh
ich
is
a
co
m
m
o
n
is
s
u
e
f
ac
ed
in
th
e
ex
tr
ac
tiv
e
ap
p
r
o
ac
h
.
Sen
te
n
ce
f
u
s
io
n
o
r
s
en
te
n
ce
co
m
p
r
e
s
s
io
n
tech
n
iq
u
es
ar
e
s
o
m
e
o
f
t
h
e
ea
r
ly
ad
o
p
tio
n
s
o
f
ab
s
tr
ac
tiv
e
m
eth
o
d
s
.
T
h
e
m
ain
is
s
u
e
f
ac
ed
b
y
th
ese
ap
p
r
o
ac
h
es
is
th
e
g
en
er
ated
h
ea
d
lin
es
w
er
e
n
o
t
ab
le
to
co
n
v
ey
th
e
m
ain
i
d
ea
o
f
t
h
e
n
e
ws an
d
wer
e
v
er
y
lack
in
g
in
g
r
am
m
atica
l stru
ctu
r
e.
T
h
e
d
ee
p
lear
n
in
g
-
b
ased
a
p
p
r
o
ac
h
with
th
e
ad
o
p
tio
n
o
f
s
eq
2
s
eq
h
as
s
h
o
wn
its
s
u
cc
ess
in
tex
t
s
u
m
m
ar
izatio
n
[
2
2
]
.
T
h
is
m
o
d
el
co
n
s
is
ts
o
f
th
e
en
co
d
er
-
d
ec
o
d
er
ar
ch
itectu
r
e
[
8
]
,
[
9
]
.
T
h
e
en
co
d
er
will
co
m
p
u
te
th
e
h
id
d
en
s
tate
o
f
ea
ch
wo
r
d
s
eq
u
en
ce
an
d
th
e
d
ec
o
d
e
r
will
ca
lcu
late
th
e
p
r
o
b
ab
ilit
y
o
f
ea
ch
wo
r
d
in
th
e
v
o
ca
b
u
lar
y
to
g
en
er
ate
th
e
o
u
t
p
u
t seq
u
en
ce
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Ma
la
y
p
h
o
n
eme
-
b
a
s
ed
s
u
b
w
o
r
d
n
ews
h
ea
d
lin
e
g
en
era
to
r
fo
r
lo
w
-
r
eso
u
r
ce
la
n
g
u
a
g
e
(
Yeo
n
g
Ts
a
n
n
P
h
u
a
)
4967
2
.
2
.
L
o
w
-
re
s
o
urce
la
ng
ua
g
e
L
R
L
ar
e
also
k
n
o
wn
as u
n
d
er
-
r
eso
u
r
ce
d
lan
g
u
a
g
es.
I
n
NL
P,
th
ese
ar
e
r
e
f
er
r
ed
as lan
g
u
ag
es
th
at
h
a
v
e
lim
ited
r
ea
d
ily
av
ailab
le
lin
g
u
is
tic
-
r
elate
d
m
ater
ials
an
d
/o
r
d
i
g
ital
d
ata.
Usu
ally
,
s
u
ch
lan
g
u
ag
es
f
ac
e
ch
allen
g
es,
in
clu
d
in
g
lim
ited
lex
ical
r
eso
u
r
ce
s
an
d
s
p
a
r
s
e
c
o
r
p
o
r
a.
I
n
th
is
r
esear
ch
,
th
e
f
o
c
u
s
is
o
n
th
e
Ma
lay
lan
g
u
ag
e.
T
h
e
Ma
lay
lan
g
u
ag
e
is
a
lan
g
u
ag
e
u
n
d
er
Nu
s
an
ta
r
a
in
th
e
Au
s
tr
o
n
esia
lan
g
u
ag
e
f
am
ily
[
2
3
]
,
[
2
4
]
.
Gen
er
ally
,
th
is
lan
g
u
ag
e
h
as
m
o
r
e
th
an
2
9
0
m
illi
o
n
n
ativ
e
s
p
ea
k
er
s
[
2
4
]
.
I
n
th
e
ar
ea
o
f
NL
P,
th
e
r
esear
ch
ac
tiv
ities
in
v
o
lv
in
g
th
e
Ma
lay
lan
g
u
ag
e
ar
e
v
er
y
lim
ited
.
I
n
t
h
e
ar
ea
o
f
tex
t
s
u
m
m
ar
izatio
n
an
d
NHG,
m
o
s
t
o
f
th
e
r
esear
ch
m
ain
ly
in
v
o
l
v
es
p
r
iv
ate
d
atasets
f
o
r
ex
p
er
im
en
ts
.
T
h
e
m
ain
s
o
u
r
ce
s
o
f
t
h
e
n
e
ws
ar
e
f
r
o
m
B
er
ita
Har
ian
[
2
5
]
,
[
2
6
]
,
Utu
s
an
Ma
lay
s
ia
[
2
5
]
,
[
2
6
]
,
an
d
B
er
n
am
a
[
2
5
]
‒
[
2
7
]
.
Gen
er
ally
,
th
er
e
ar
e
n
o
r
ea
d
ily
p
u
b
lic
d
atasets
av
ailab
le.
2
.
3
.
Rec
urre
nt
neura
l net
wo
rk
As
a
class
o
f
ar
tific
ial
n
eu
r
al
n
etwo
r
k
s
(
ANN)
,
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
(
R
NNs)
ar
e
ca
p
ab
le
o
f
p
r
o
ce
s
s
in
g
s
eq
u
en
tial
d
ata
b
y
m
ain
tain
in
g
a
h
id
d
e
n
s
tate
th
a
t
ca
p
tu
r
es
in
f
o
r
m
atio
n
f
r
o
m
p
r
ev
io
u
s
in
p
u
ts
.
T
h
e
n
atu
r
e
o
f
th
is
d
esig
n
m
a
k
es
R
NN
s
u
itab
le
f
o
r
s
eq
u
en
tial
tas
k
s
in
tex
t
g
en
er
atio
n
[
2
8
]
.
T
h
er
ef
o
r
e,
in
s
eq
2
s
eq
ar
ch
itectu
r
es,
R
NNs
ar
e
u
s
ed
i
n
b
o
th
e
n
co
d
e
r
an
d
d
ec
o
d
e
r
to
ca
p
tu
r
e
t
h
e
s
eq
u
e
n
tial
in
f
o
r
m
a
tio
n
f
r
o
m
i
n
p
u
t
an
d
g
en
er
ate
o
u
tp
u
t in
s
eq
u
en
ce
.
T
h
e
co
m
m
o
n
R
NN
ar
ch
itectu
r
es in
clu
d
e
b
asic RNN,
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
,
b
i
d
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
B
iLST
M)
,
an
d
g
ated
r
ec
u
r
r
e
n
t
u
n
it
(
GR
U)
.
T
h
e
b
asic
R
N
N
is
also
ca
lled
v
an
illa
R
NN.
I
t
wo
r
k
s
b
y
s
to
r
in
g
an
d
u
p
d
atin
g
t
h
e
h
i
d
d
en
s
tate
at
ea
ch
tim
e
s
tep
.
Ho
wev
er
,
it
h
as
a
wea
k
n
ess
in
h
an
d
lin
g
lo
n
g
-
ter
m
d
e
p
en
d
e
n
cies
[
2
9
]
.
T
h
e
L
STM
is
co
n
s
i
d
er
ed
an
ad
v
a
n
ce
d
v
ar
ian
t
o
f
v
an
illa
R
NN.
I
t
is
d
esig
n
ed
to
o
v
er
c
o
m
e
th
e
is
s
u
e
o
f
lo
n
g
-
ter
m
d
ep
e
n
d
en
cies
[
3
0
]
.
T
h
e
L
STM
ce
ll
h
as
th
r
ee
g
ates,
th
e
in
p
u
t,
o
u
tp
u
t,
an
d
f
o
r
g
et
g
ates.
T
h
ese
g
ates
will
m
an
ag
e
th
e
f
lo
w
o
f
th
e
in
f
o
r
m
atio
n
f
r
o
m
ea
ch
tim
e
s
tep
.
I
t
ca
n
ca
p
tu
r
e
lo
n
g
d
ep
en
d
en
cies
an
d
m
ain
t
ain
in
f
o
r
m
atio
n
o
v
er
lo
n
g
er
s
eq
u
en
ce
s
.
T
h
e
Bi
L
STM
tak
e
s
a
s
tep
f
u
r
th
e
r
b
y
m
er
g
in
g
in
f
o
r
m
atio
n
f
r
o
m
th
e
p
ast
an
d
th
e
f
u
tu
r
e
in
ea
c
h
tim
e
s
tep
[
3
1
]
.
T
h
is
R
NN
is
ab
le
to
h
o
ld
th
e
in
f
o
r
m
atio
n
o
f
th
e
p
ast
an
d
f
u
tu
r
e
u
s
in
g
th
e
f
o
r
war
d
s
tate
in
a
p
o
s
itiv
e
tim
e
d
ir
ec
tio
n
an
d
t
h
e
b
ac
k
wa
r
d
s
tate
in
a
n
eg
ativ
e
tim
e
d
ir
ec
tio
n
.
R
esear
ch
b
y
C
h
o
et
a
l
.
[
3
2
]
,
th
e
G
R
U
was
p
r
o
p
o
s
ed
.
T
h
is
R
NN
i
s
an
o
th
er
v
ar
ian
t
o
f
th
e
v
an
illa
R
NN.
I
t
s
im
p
lifie
s
th
e
g
ate
d
esig
n
b
y
u
s
in
g
th
e
r
eset
an
d
u
p
d
ate
g
ates,
wh
ich
will
d
ec
id
e
th
e
am
o
u
n
t
o
f
in
f
o
r
m
atio
n
t
o
k
ee
p
o
r
d
is
ca
r
d
f
r
o
m
p
r
ev
io
u
s
an
d
cu
r
r
en
t
tim
e
s
tep
s
.
2
.
4
.
Su
bwo
rd
t
o
k
eniza
t
io
n
I
n
NL
P
-
r
elate
d
task
s
,
wo
r
d
r
e
p
r
esen
tatio
n
,
also
k
n
o
wn
as wo
r
d
em
b
e
d
d
in
g
,
p
lay
s
a
cr
itica
l r
o
le.
T
h
is
p
r
o
ce
s
s
co
n
v
er
ts
th
e
in
p
u
t
tex
t
in
to
n
u
m
er
ical
v
ec
t
o
r
s
,
wh
ich
co
n
tain
th
e
s
em
an
tic
r
ep
r
esen
tatio
n
b
etwe
en
th
e
wo
r
d
s
an
d
th
e
c
o
n
tex
t
in
f
o
r
m
atio
n
o
f
th
e
te
x
t.
T
h
e
tr
ad
it
io
n
al
wo
r
d
r
ep
r
esen
tatio
n
tec
h
n
iq
u
e
u
s
es
o
n
e
-
h
o
t
en
co
d
in
g
wh
ich
in
v
o
lv
es
a
1
in
th
e
co
r
r
esp
o
n
d
in
g
wo
r
d
p
o
s
itio
n
’
s
in
d
ex
a
n
d
0
f
o
r
t
h
e
r
est
o
f
th
e
wo
r
d
s
’
in
d
ices.
T
h
is
r
ep
r
esen
tatio
n
lead
s
to
s
p
ar
s
e
m
atr
ix
f
o
r
m
atio
n
in
th
e
e
m
b
ed
d
in
g
wh
ich
is
n
o
t
co
m
p
u
tatio
n
ally
o
p
tim
ized
[
3
3
]
.
B
esid
es th
at,
th
is
r
ep
r
ese
n
tatio
n
is
n
o
t c
ap
a
b
le
o
f
ca
p
tu
r
in
g
th
e
s
em
an
tic
r
elatio
n
s
h
ip
b
etwe
en
wo
r
d
s
.
Me
an
wh
ile,
th
e
o
th
er
class
ic
ty
p
es
o
f
em
b
ed
d
in
g
s
ad
o
p
t
s
tatis
tical
p
r
o
b
ab
ilit
y
ap
p
r
o
ac
h
es
s
u
ch
as
W
o
r
d
2
Vec
[
3
3
]
an
d
Glo
Ve
[
3
4
]
.
B
o
th
ap
p
r
o
ac
h
es
ar
e
a
b
l
e
to
ca
p
tu
r
e
t
h
e
wo
r
d
r
elatio
n
s
h
ip
an
d
u
s
e
t
h
e
n
eig
h
b
o
r
in
g
wo
r
d
s
t
o
f
o
r
m
m
e
an
in
g
f
u
l
wo
r
d
r
elatio
n
s
h
ip
s
.
Ho
wev
er
,
t
h
ese
two
ap
p
r
o
ac
h
es a
r
e
n
o
t
ca
p
ab
le
o
f
h
an
d
lin
g
n
ew
wo
r
d
s
o
r
u
n
s
ee
n
wo
r
d
s
f
r
o
m
th
e
in
p
u
t
o
f
th
e
NL
P
task
s
.
B
esid
es
th
at,
th
ey
n
ee
d
a
lar
g
e
em
b
ed
d
in
g
f
o
r
lan
g
u
a
g
es
with
co
m
p
lex
m
o
r
p
h
o
lo
g
ical
s
tr
u
ctu
r
es.
T
o
o
v
er
c
o
m
e
th
e
is
s
u
es
o
f
r
ar
e
wo
r
d
s
an
d
OOV,
s
u
b
wo
r
d
em
b
ed
d
in
g
is
in
tr
o
d
u
ce
d
.
T
h
is
em
b
e
d
d
in
g
will
tak
e
an
in
p
u
t
wo
r
d
a
n
d
s
p
lit
th
e
wo
r
d
in
to
s
m
aller
u
n
its
s
u
ch
as
ch
a
r
ac
te
r
n
-
g
r
am
o
r
lin
g
u
is
tic
s
u
b
wo
r
d
u
n
its
.
So
m
e
o
f
th
e
n
o
tab
le
wo
r
k
s
in
m
o
r
p
h
em
e
s
eg
m
en
tatio
n
ar
e
in
[
3
5
]
‒
[
3
8
]
.
3.
P
RO
P
O
SE
D
M
E
T
H
O
D
T
h
e
p
r
o
p
o
s
ed
n
ew
MSAS
m
eth
o
d
co
m
p
r
is
es
o
f
a
h
y
b
r
id
o
f
th
e
e
x
tr
ac
tiv
e
a
n
d
ab
s
tr
ac
tiv
e
s
u
m
m
ar
izatio
n
m
o
d
els.
T
h
e
M
SAS
is
co
n
s
tr
u
cted
with
f
o
u
r
m
ajo
r
alg
o
r
ith
m
ic
elem
e
n
ts
,
a
s
s
h
o
wn
in
Fig
u
r
e
1
.
T
h
ese
f
o
u
r
alg
o
r
ith
m
ic
elem
en
ts
ar
e
:
i
)
lead
in
g
s
en
ten
ce
n
ews
T
ex
tR
an
k
;
ii
)
Ma
lay
p
h
o
n
e
m
e
s
u
b
wo
r
d
to
k
en
izer
,
iii
)
Ma
lay
p
h
o
n
em
e
s
u
b
wo
r
d
to
k
e
n
izer
s
u
b
wo
r
d
e
m
b
ed
d
in
g
;
an
d
iv
)
s
eq
2
s
eq
en
co
d
er
a
n
d
d
ec
o
d
er
.
T
h
e
MSAS
m
o
d
el
h
as
a
d
o
p
te
d
a
h
y
b
r
id
d
esig
n
co
n
ce
p
t
b
y
b
in
d
in
g
t
h
e
e
x
tr
ac
tiv
e
a
n
d
ab
s
tr
ac
tiv
e
s
u
m
m
ar
izatio
n
ap
p
r
o
ac
h
es
in
to
o
n
e.
T
h
e
m
o
tiv
atio
n
f
o
r
u
s
in
g
th
is
h
y
b
r
id
d
e
s
ig
n
is
to
o
v
e
r
co
m
e
t
h
e
co
m
m
o
n
is
s
u
e
f
ac
ed
in
th
e
g
en
er
ic
s
eq
2
s
eq
tr
ai
n
in
g
m
o
d
el
wh
en
co
m
p
u
tin
g
with
l
o
n
g
in
p
u
t
s
eq
u
en
ce
s
.
As
th
e
tr
ain
in
g
s
eq
u
en
ce
len
g
th
in
cr
ea
s
es,
th
e
h
ar
d
war
e
m
e
m
o
r
y
r
eq
u
ir
em
en
t in
c
r
ea
s
es.
T
h
er
ef
o
r
e,
th
is
h
y
b
r
id
a
p
p
r
o
a
ch
ai
m
s
to
o
v
er
co
m
e
th
e
h
ig
h
m
em
o
r
y
r
eq
u
ir
e
m
en
t in
t
h
e
m
o
d
el
tr
ai
n
in
g
.
I
n
th
e
f
ir
s
t
co
m
p
u
tatio
n
al
ele
m
en
t
o
f
MSAS,
th
e
e
x
tr
ac
tio
n
ap
p
r
o
ac
h
h
as
b
ee
n
in
tr
o
d
u
ce
d
to
r
e
d
u
ce
th
e
o
v
er
all
le
n
g
th
o
f
th
e
in
p
u
t seq
u
en
ce
to
t
h
e
tr
ain
in
g
u
s
in
g
lead
in
g
s
en
ten
ce
n
ews
T
ex
tR
an
k
,
wh
ich
h
elp
s
to
r
ed
u
ce
t
h
e
in
p
u
t
b
y
e
x
tr
ac
tin
g
s
alien
t
n
ews
s
en
ten
ce
s
f
r
o
m
t
h
e
in
p
u
t
s
o
u
r
ce
[
3
9
]
.
Su
b
s
eq
u
en
tly
,
th
e
ex
tr
ac
te
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
9
6
5
-
4
9
7
5
4968
s
alien
t
n
ews
s
en
ten
ce
s
’
wo
r
d
s
in
to
s
u
b
wo
r
d
t
o
k
en
s
ar
e
t
o
k
en
i
ze
d
u
s
in
g
th
e
Ma
lay
p
h
o
n
em
e
s
u
b
wo
r
d
to
k
e
n
ize
r
.
Fin
ally
,
th
e
to
k
en
ized
s
u
b
wo
r
d
s
ar
e
f
ed
in
to
th
e
s
eq
2
s
eq
m
o
d
el
an
d
tr
ain
ed
with
th
e
Ma
lay
p
h
o
n
em
e
s
u
b
wo
r
d
em
b
ed
d
in
g
.
Fig
u
r
e
1
.
MSAS
3
.
1
.
L
ea
din
g
s
ent
ence
news
T
ex
t
Ra
nk
T
h
e
T
ex
tR
an
k
alg
o
r
ith
m
is
a
n
u
n
s
u
p
e
r
v
is
ed
lear
n
i
n
g
alg
o
r
ith
m
.
I
n
th
e
b
e
g
in
n
in
g
,
th
is
alg
o
r
ith
m
r
eq
u
ir
es th
e
in
p
u
t te
x
t to
b
e
tr
a
n
s
f
o
r
m
ed
i
n
to
a
f
o
r
m
o
f
v
ec
to
r
r
ep
r
esen
tatio
n
u
s
in
g
th
e
ter
m
f
r
eq
u
e
n
cy
-
in
v
er
s
e
d
o
cu
m
e
n
t
f
r
eq
u
e
n
cy
(
T
F
-
I
DF)
r
ep
r
esen
tatio
n
at
th
e
s
en
ten
ce
p
r
e
-
p
r
o
ce
s
s
in
g
lev
el.
T
h
ese
v
ec
to
r
r
ep
r
esen
tatio
n
s
will
b
e
f
ed
in
to
th
e
T
ex
tR
an
k
alg
o
r
ith
m
to
co
n
s
tr
u
ct
th
e
r
ela
tio
n
b
etwe
en
th
e
n
o
d
es.
I
n
o
r
d
er
to
allo
w
all
th
e
n
o
d
es
with
eq
u
al
p
r
o
b
a
b
ilit
y
to
b
e
s
elec
ted
as
th
e
s
tar
tin
g
n
o
d
e
at
ea
ch
iter
atio
n
,
th
e
T
ex
tR
an
k
alg
o
r
ith
m
ap
p
lies
r
an
d
o
m
p
r
o
b
ab
ilit
y
to
in
itialize
all
n
o
d
es
at
th
e
b
eg
in
n
in
g
o
f
ea
ch
lear
n
in
g
p
r
o
ce
s
s
.
T
h
e
r
e
latio
n
s
h
ip
b
etwe
en
th
e
n
o
d
es
is
r
e
p
r
esen
ted
b
y
th
e
“stre
n
g
th
”
b
etwe
en
th
e
n
o
d
e
s
.
T
h
e
s
tr
en
g
th
is
c
o
m
p
u
ted
b
ased
o
n
th
e
wo
r
d
s
’
s
im
ilar
ities
[
39
].
T
h
e
lead
in
g
s
en
ten
ce
n
ews
T
ex
tR
an
k
ad
o
p
ted
th
e
T
ex
tR
an
k
ap
p
r
o
ac
h
th
at
is
cu
s
to
m
ized
f
o
r
n
ews
ar
ticle
s
alien
t
s
en
ten
ce
s
elec
tio
n
.
Acc
o
r
d
in
g
to
T
s
an
n
et
a
l
.
[
3
9
]
th
is
alg
o
r
ith
m
will
r
eq
u
ir
e
s
ettin
g
1
to
th
e
weig
h
t o
f
n
o
d
e
o
f
th
e
f
ir
s
t sen
ten
ce
o
f
th
e
in
p
u
t te
x
t.
T
h
e
(
1
)
is
u
s
ed
to
ex
p
r
ess
th
e
weig
h
t
o
f
a
n
o
d
e
.
(
)
=
(
1
−
)
+
∗
∑
∑
∈
(
)
∈
(
)
(
)
(
1
)
I
n
th
e
eq
u
atio
n
,
t
h
e
weig
h
t b
et
wee
n
th
e
n
o
d
es is
ex
p
r
ess
ed
a
s
an
d
th
e
d
a
m
p
in
g
f
ac
to
r
,
is
s
et
to
0
.
8
5
.
3
.
2
.
M
a
la
y
ph
o
nem
e
s
ub
wo
rd
t
o
k
enizer
T
h
e
Ma
lay
lan
g
u
ag
e
is
a
t
y
p
e
o
f
ag
g
lu
tin
ativ
e
la
n
g
u
a
g
e.
T
h
is
lan
g
u
a
g
e
is
f
u
ll
o
f
m
o
r
p
h
em
es.
An
ex
am
p
le
o
f
a
r
o
o
t
wo
r
d
,
“jan
g
k
a”
wh
ich
m
ea
n
s
“e
x
p
ec
t”
in
E
n
g
lis
h
.
T
h
e
ap
p
licatio
n
o
f
a
p
r
ef
ix
will
b
ec
o
m
e
“m
en
jan
g
k
a”
wh
ich
m
ea
n
s
“to
ex
p
ec
t”
in
E
n
g
lis
h
.
T
h
e
ap
p
lic
atio
n
o
f
a
s
u
f
f
ix
will
tu
r
n
t
h
e
w
o
r
d
in
t
o
“jan
g
k
aa
n
”
wh
ich
m
ea
n
s
“e
x
p
ec
tatio
n
s
”
i
n
E
n
g
lis
h
.
T
h
e
ap
p
licatio
n
o
f
t
h
e
cir
cu
m
f
ix
ca
n
tu
r
n
t
h
e
wo
r
d
in
to
“d
ijan
g
k
a
k
an
”
wh
ich
m
ea
n
s
“e
x
p
ec
ted
”
in
E
n
g
lis
h
.
T
h
e
ap
p
licatio
n
o
f
v
ar
io
u
s
p
h
o
n
em
es
h
as
ch
a
n
g
ed
th
e
r
o
o
t
wo
r
d
in
to
d
if
f
er
en
t m
ea
n
in
g
s
in
t
h
e
tar
g
e
t tr
an
s
lated
lan
g
u
ag
e
[4
0
]
.
T
h
e
ap
p
licatio
n
o
f
tr
ad
itio
n
al
wo
r
d
em
b
ed
d
in
g
will
r
eq
u
ir
e
th
e
r
ep
r
esen
tatio
n
o
f
a
s
in
g
le
wo
r
d
th
at
h
as
m
u
ltip
le
d
if
f
er
e
n
t
ap
p
licat
io
n
s
o
f
m
o
r
p
h
e
m
es.
T
h
is
will
in
cr
ea
s
e
th
e
n
u
m
b
er
o
f
wo
r
d
s
in
th
e
em
b
ed
d
in
g
s
p
ac
e.
T
h
e
ap
p
licatio
n
o
f
s
tem
m
in
g
is
n
o
t
a
p
r
ac
tical
ap
p
r
o
ac
h
as
th
e
co
n
tex
t
o
f
a
s
p
ec
if
ic
tr
an
s
lated
wo
r
d
m
ay
b
e
d
if
f
er
e
n
t
in
m
ea
n
i
n
g
.
B
esid
es
th
at,
r
ed
u
cin
g
th
e
n
u
m
b
e
r
o
f
wo
r
d
s
in
t
h
e
wo
r
d
em
b
ed
d
in
g
s
p
ac
e
will
also
lead
to
OOV
as so
m
e
o
f
th
e
w
o
r
d
s
will b
e
r
em
o
v
ed
d
u
r
in
g
t
h
e
wo
r
d
r
ep
r
esen
tatio
n
p
r
o
ce
s
s
.
T
o
o
v
er
co
m
e
th
e
co
m
p
le
x
m
o
r
p
h
o
lo
g
ical
s
tr
u
ctu
r
e
in
th
e
M
alay
lan
g
u
ag
e
c
o
m
p
a
r
ed
t
o
th
e
E
n
g
lis
h
lan
g
u
ag
e,
th
is
s
tu
d
y
p
r
o
p
o
s
es
a
n
ew
Ma
lay
p
h
o
n
em
e
-
b
ase
d
to
k
en
izatio
n
ap
p
r
o
ac
h
.
T
h
i
s
n
ew
to
k
en
izatio
n
ap
p
r
o
ac
h
will
s
p
lit
a
wo
r
d
in
t
o
m
u
ltip
le
s
u
b
wo
r
d
to
k
en
s
.
I
n
s
t
ea
d
o
f
alg
o
r
ith
m
-
b
ased
to
k
e
n
izatio
n
,
th
is
ap
p
r
o
ac
h
will
s
p
lit
a
Ma
lay
wo
r
d
in
to
s
u
b
wo
r
d
to
k
e
n
s
b
ased
o
n
th
e
f
o
u
r
d
is
tin
ct
s
y
llab
le
s
tr
u
ctu
r
es
in
th
e
Ma
lay
lan
g
u
ag
e
[4
1
]
.
T
h
ese
s
y
llab
les
ar
e
b
ased
o
n
th
e
v
o
wel,
V,
an
d
co
n
s
o
n
an
t,
C
,
th
e
n
th
e
f
o
u
r
p
r
o
p
o
s
ed
s
y
llab
le
s
tr
u
ctu
r
e
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Ma
la
y
p
h
o
n
eme
-
b
a
s
ed
s
u
b
w
o
r
d
n
ews
h
ea
d
lin
e
g
en
era
to
r
fo
r
lo
w
-
r
eso
u
r
ce
la
n
g
u
a
g
e
(
Yeo
n
g
Ts
a
n
n
P
h
u
a
)
4969
ar
e
s
u
b
s
eq
u
en
tly
:
V,
C
V,
VC
,
an
d
C
VC
to
b
e
u
s
ed
t
o
d
eter
m
in
e
th
e
s
u
b
wo
r
d
to
k
en
s
.
T
ab
le
1
s
h
o
ws
a
s
et
o
f
s
am
p
les o
f
Ma
lay
wo
r
d
s
th
at
a
r
e
b
ased
o
n
v
ar
io
u
s
s
y
llab
ic
s
tr
u
ctu
r
es a
s
d
escr
ib
ed
a
b
o
v
e
.
T
h
e
p
r
o
p
o
s
ed
n
ew
to
k
en
izer
will
p
er
f
o
r
m
th
e
to
k
en
izatio
n
p
r
o
ce
s
s
to
s
tr
u
ctu
r
e
th
e
in
p
u
t
tex
t
in
to
Ma
lay
p
h
o
n
em
e
s
u
b
wo
r
d
s
th
r
o
u
g
h
an
iter
ativ
e
p
r
o
ce
s
s
f
o
r
ea
c
h
ch
ar
ac
ter
i
n
th
e
in
p
u
t
tex
t
u
n
til
all
th
e
ch
ar
ac
ter
s
in
ea
ch
wo
r
d
h
av
e
c
o
m
p
leted
f
o
r
m
in
g
th
e
co
m
m
o
n
s
y
llab
le
s
tr
u
ctu
r
e.
A
s
p
ec
ial
en
d
-
of
-
wo
r
d
s
y
m
b
o
l,
‘
_
,
’
will
b
e
ad
d
ed
to
th
e
last
s
u
b
wo
r
d
to
k
en
to
in
d
icate
th
e
e
n
d
o
f
a
wo
r
d
.
T
h
u
s
,
th
e
n
ew
to
k
e
n
izatio
n
p
r
o
ce
s
s
is
co
m
p
leted
.
T
ab
le
1
.
Sy
llab
le
s
tr
u
ctu
r
es
o
f
Ma
lay
wo
r
d
s
Ty
p
e
o
f
s
y
l
l
a
b
l
e
S
y
l
l
a
b
l
e
st
r
u
c
t
u
r
e
Ex
a
m
p
l
e
w
o
r
d
M
o
n
o
sy
l
l
a
b
i
c
C
V
C
CV
c
i
k
,
s
u
p
,
J
u
n
d
i
,
k
e
D
i
ssy
l
l
a
b
i
c
C
V
+
C
V
V
+
C
V
V
+
C
V
C
p
e
l
i
,
c
a
ri
,
l
a
g
u
a
b
u
,
i
t
u
,
i
n
i
u
l
a
t
,
i
k
a
t
Tr
i
sy
l
l
a
b
i
c
C
V
+
C
V
+
C
V
C
V
+
V
+
C
V
C
V
+
C
V
+
V
p
e
r
a
s
a
,
k
e
r
u
si
b
u
a
y
a
,
k
u
a
sa
semu
a
,
k
e
t
u
a
Te
t
r
a
s
y
l
l
a
b
i
c
C
V
+
C
V
+
C
V
+
C
V
C
V
+
C
V
C
+
C
V
+
V
C
V
+
C
V
+
C
V
+
CV
d
a
r
i
p
a
d
a
p
e
r
e
m
p
u
a
n
a
p
a
b
i
l
a
3
.
3
.
Sequ
ence
-
to
-
s
equence
I
n
2
0
1
4
,
Sh
war
tz
an
d
Z
h
a
n
g
[4
2
]
p
r
o
p
o
s
ed
th
e
s
eq
2
s
eq
m
o
d
el
wh
ich
c
o
n
s
is
ts
o
f
an
e
n
c
o
d
er
a
n
d
a
d
ec
o
d
er
.
T
h
e
en
co
d
e
r
co
n
s
is
ts
o
f
a
s
tack
o
f
R
NN
-
b
ased
ce
ll
s
.
T
h
e
in
p
u
t
s
eq
u
en
ce
is
f
ed
in
to
th
e
en
co
d
er
.
T
h
en
,
th
e
in
p
u
t
v
ec
t
o
r
will
b
e
c
o
m
p
u
ted
,
a
n
d
th
e
in
te
r
n
al
s
tate
wil
l
b
e
s
to
r
e
d
in
th
e
R
N
N
-
b
ased
ce
ll.
T
h
e
(
2
)
s
h
o
ws
th
e
co
m
p
u
tatio
n
o
f
th
e
h
i
d
d
en
s
tate
ℎ
is
co
m
p
u
ted
u
s
in
g
th
e
w
eig
h
t
f
u
n
ctio
n
o
f
th
e
e
x
is
tin
g
in
p
u
t
,
p
r
ev
io
u
s
s
tate
ℎ
−
1
an
d
t
h
e
n
etwo
r
k
weig
h
t,
.
ℎ
=
(
ℎℎ
ℎ
−
1
+
ℎ
)
(
2
)
On
th
e
o
th
er
s
id
e
o
f
th
e
m
o
d
el,
th
e
d
ec
o
d
er
is
f
o
r
m
ed
u
s
i
n
g
a
s
tack
o
f
r
ec
u
r
r
en
t
-
b
ased
ce
lls
.
T
h
e
d
ec
o
d
er
will
tak
e
in
th
e
f
in
al
s
tate
o
f
th
e
en
co
d
er
as
in
itial
in
p
u
t.
E
ac
h
tim
e
s
tep
,
th
e
h
id
d
en
s
tate
ℎ
is
ca
lcu
lated
u
s
in
g
th
e
o
u
tp
u
t
f
r
o
m
th
e
p
r
e
v
io
u
s
tim
e
s
tep
u
s
in
g
(
3
)
.
ℎ
=
(
ℎℎ
ℎ
−
1
)
(
3
)
T
h
e
o
u
t
p
u
t
o
f
ea
ch
tim
e
s
tep
,
is
co
m
p
u
ted
u
s
in
g
a
So
f
tm
ax
f
u
n
ctio
n
with
th
e
co
m
b
i
n
atio
n
o
f
th
e
h
id
d
en
s
tate
o
f
th
e
cu
r
r
en
t
tim
e
s
tep
an
d
th
e
r
esp
ec
tiv
e
weig
h
t
.
T
h
e
p
r
o
b
ab
ilit
y
v
ec
to
r
o
f
t
h
e
o
u
tp
u
t
is
s
h
o
wn
in
(
4
)
.
=
(
ℎ
)
(
4
)
I
n
th
e
p
r
o
p
o
s
ed
s
eq
2
s
eq
m
o
d
el,
a
s
in
g
le
-
lay
er
R
NN
-
b
ased
ce
ll
is
u
s
ed
as
an
en
c
o
d
er
,
an
d
a
s
in
g
le
-
lay
er
R
NN
-
b
ased
ce
ll a
s
th
e
d
ec
o
d
e
r
.
3
.
4
.
At
t
ent
io
n
m
ec
ha
nis
m
I
n
2
0
1
5
,
B
ah
d
a
n
au
et
a
l.
[
7
]
f
ir
s
t
in
tr
o
d
u
ce
d
th
e
atten
tio
n
m
ec
h
an
is
m
in
to
th
e
s
eq
2
s
eq
m
o
d
el.
T
h
e
im
p
lem
en
tatio
n
will
o
v
er
c
o
m
e
th
e
lim
itatio
n
s
o
f
tr
ad
itio
n
al
s
eq
2
s
eq
in
h
a
n
d
lin
g
lo
n
g
s
eq
u
en
ce
s
.
I
n
th
e
tr
ad
itio
n
al
s
eq
2
s
eq
,
th
e
R
NN
-
b
ased
ce
lls
m
u
s
t
m
ain
tain
all
th
e
h
id
d
en
s
tates
f
o
r
ea
ch
tim
e
s
tep
o
f
th
e
in
p
u
t
s
eq
u
en
ce
u
n
til
th
e
en
d
o
f
th
e
in
p
u
t
s
eq
u
en
ce
.
Hen
ce
,
th
is
in
f
o
r
m
atio
n
r
ete
n
tio
n
lea
d
s
to
h
ig
h
m
em
o
r
y
an
d
co
m
p
u
tatio
n
al
r
e
q
u
ir
em
e
n
ts
.
T
h
is
b
ec
o
m
es
ev
en
m
o
r
e
ch
allen
g
in
g
in
h
an
d
lin
g
all
th
e
h
id
d
en
s
tates
o
f
lo
n
g
-
r
a
n
g
e
d
ep
en
d
en
cies.
T
h
e
atten
tio
n
m
ec
h
a
n
is
m
will
allo
w
th
e
m
o
d
el
t
o
o
n
ly
f
o
cu
s
o
n
t
h
e
r
ele
v
an
t
p
ar
ts
o
f
th
e
in
p
u
t
ef
f
ec
tiv
ely
with
o
u
t
in
c
r
ea
s
in
g
th
e
m
em
o
r
y
a
n
d
co
m
p
u
tatio
n
a
l
r
eq
u
ir
e
m
en
ts
[4
3
]
.
T
h
e
d
ec
o
d
er
will
s
elec
tiv
ely
f
o
cu
s
o
n
th
e
d
if
f
er
e
n
t
p
ar
ts
o
f
t
h
e
in
p
u
t
s
eq
u
en
ce
d
u
r
i
n
g
ea
c
h
d
ec
o
d
in
g
s
tep
b
ased
o
n
t
h
e
att
en
tio
n
weig
h
t
[4
4
]
.
W
ith
th
e
im
p
lem
en
tatio
n
o
f
th
e
atten
tio
n
m
ec
h
an
is
m
,
t
h
e
s
eq
2
s
eq
m
o
d
el
will
co
n
s
is
t
o
f
th
r
e
e
p
ar
ts
:
th
e
e
n
co
d
e
r
,
th
e
d
ec
o
d
er
,
a
n
d
th
e
atten
tio
n
lay
er
.
T
h
e
atten
tio
n
lay
er
co
n
tain
s
th
r
ee
p
ar
ts
th
at
i
n
clu
d
e
th
e
alig
n
m
en
t
lay
er
,
atten
tio
n
weig
h
ts
,
an
d
c
o
n
tex
t
v
ec
to
r
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
9
6
5
-
4
9
7
5
4970
I
n
th
e
atten
tio
n
m
ec
h
a
n
is
m
,
th
e
alig
n
m
en
t
lay
e
r
p
lay
s
a
n
im
p
o
r
tan
t
r
o
le
in
c
o
m
p
u
tin
g
th
e
alig
n
m
en
t
s
co
r
es
b
etwe
en
th
e
in
p
u
t
s
eq
u
e
n
ce
an
d
th
e
o
u
tp
u
t
s
eq
u
en
ce
o
f
th
e
s
eq
2
s
eq
.
T
h
ese
s
co
r
es
ar
e
a
ls
o
ca
lled
atten
tio
n
s
co
r
es
o
r
atten
tio
n
weig
h
ts
.
T
h
e
am
o
u
n
t
o
f
t
h
ese
s
co
r
es
d
eter
m
in
es
h
o
w
m
u
ch
t
h
e
f
o
c
u
s
is
o
n
a
p
ar
ticu
lar
in
p
u
t
s
eq
u
en
ce
wh
en
g
en
e
r
atin
g
a
s
p
ec
if
ic
o
u
tp
u
t
s
eq
u
e
n
ce
.
T
h
e
c
u
r
r
en
t
s
tate
o
f
th
e
alig
n
m
en
t
s
co
r
e
ca
lcu
latio
n
will
in
v
o
lv
e
th
e
h
y
p
er
b
o
lic
tan
g
en
t
f
u
n
ctio
n
th
at
ap
p
lies
to
th
e
s
co
r
es
o
f
th
e
p
r
ev
i
o
u
s
s
tate
ℎ
−
1
an
d
p
r
ev
io
u
s
s
tate
−
1
as sh
o
wn
in
(
5
)
.
=
ℎ
(
−
1
+
ℎℎ
ℎ
−
1
)
(
5
)
I
n
th
e
p
r
o
p
o
s
ed
m
o
d
el,
th
e
o
r
i
g
in
al
h
y
p
er
b
o
lic
tan
g
en
t
f
u
n
cti
o
n
is
r
ep
lace
d
with
a
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U
)
f
u
n
ctio
n
as
s
h
o
wn
in
(
6
)
.
T
h
is
im
p
lem
en
tatio
n
aim
s
to
s
o
lv
e
th
e
v
an
is
h
in
g
g
r
a
d
i
en
t
th
at
co
m
m
o
n
ly
o
cc
u
r
s
in
u
s
in
g
th
e
tan
h
f
u
n
ctio
n
[4
5
]
‒
[4
7
]
.
B
esid
es th
at,
it will a
llo
w
m
o
r
e
lo
ca
lized
atte
n
tio
n
.
=
(
−
1
+
ℎℎ
ℎ
−
1
)
(
6
)
T
h
e
atten
tio
n
weig
h
ts
f
o
r
a
tar
g
et
wo
r
d
a
n
d
a
s
et
o
f
s
o
u
r
ce
w
o
r
d
s
ar
e
s
h
o
wn
in
(
7
)
.
=
ex
p
(
)
∑
ex
p
(
)
|
|
=
1
(
7
)
T
h
e
atten
tio
n
weig
h
t
an
d
th
e
h
id
d
en
s
tate
ℎ
ar
e
u
s
ed
t
o
co
m
p
u
te
t
h
e
co
n
te
x
t
v
ec
to
r
as
s
h
o
wn
in
(
8
)
.
T
h
ese
weig
h
ts
ar
e
o
b
tain
e
d
b
y
m
u
ltip
ly
in
g
t
h
e
en
co
d
er
s
tates a
n
d
th
e
d
ec
o
d
e
r
h
id
d
en
s
tates.
=
∑
ℎ
|
|
=
1
(
8
)
T
h
e
f
in
al
d
is
tr
ib
u
tio
n
p
r
o
b
ab
il
ity
is
ca
lcu
lated
b
y
a
p
p
ly
in
g
a
s
o
f
tm
ax
f
u
n
ctio
n
as sh
o
wn
in
(
9
)
.
=
(
)
(
9
)
4.
E
XP
E
R
I
M
E
N
T
AND
R
E
SU
L
T
S
T
wo
R
NN
-
b
ased
s
eq
2
s
eq
m
o
d
els,
th
e
L
STM
an
d
Bi
L
S
T
M
s
eq
2
s
eq
m
o
d
els
ar
e
u
s
ed
as
th
e
ex
p
er
im
en
t’
s
b
aselin
e
m
o
d
els to
test
ag
ain
s
t
t
h
e
n
ew
MSAS
-
tan
h
a
n
d
MSAS
-
R
eL
U
m
o
d
els.
All
th
ese
m
o
d
els
wer
e
co
n
f
ig
u
r
e
d
with
th
e
B
a
h
d
an
au
atten
tio
n
m
ec
h
a
n
is
m
.
T
h
is
s
ec
tio
n
d
etails
th
e
d
at
aset
s
elec
tio
n
,
tex
t
clea
n
in
g
,
in
p
u
t p
r
e
-
p
r
o
ce
s
s
in
g
,
an
d
th
e
m
o
d
el
c
o
n
f
ig
u
r
atio
n
s
o
f
th
e
e
x
p
er
im
e
n
t.
4
.
1
.
Da
t
a
s
et
T
h
e
d
ataset
u
s
ed
in
th
is
p
ap
er
is
o
n
lin
e
Ma
lay
lan
g
u
ag
e
n
ews
th
at
ar
e
r
an
d
o
m
l
y
co
llected
f
r
o
m
v
ar
io
u
s
n
ews
web
s
ites
(
i.e
.
B
er
n
am
a.
co
m
,
m
y
Me
tr
o
,
an
d
th
e
m
a
lay
s
ian
in
s
ig
h
t.c
o
m
)
.
T
h
ese
n
ews
ar
ticles
wer
e
p
u
b
lis
h
ed
b
etwe
en
2
0
1
7
an
d
2
0
1
9
.
Ab
o
u
t 9
8
.
9
% o
f
th
is
n
ews
ar
e
h
ar
d
n
ews o
n
cu
r
r
e
n
t a
f
f
air
s
.
T
h
e
r
em
ain
in
g
n
ews
ar
e
h
u
m
an
in
ter
est
s
o
f
t
n
ews.
T
h
e
b
r
ea
k
d
o
wn
o
f
th
e
n
ews
g
en
r
es
is
as
m
en
tio
n
ed
:
i)
n
atio
n
al
n
ews;
ii)
s
p
o
r
ts
n
ews;
iii)
ec
o
n
o
m
y
a
n
d
f
in
a
n
ce
n
ews;
iv
)
wo
r
ld
n
e
ws;
v
i)
en
ter
tain
m
en
t
a
n
d
ce
leb
r
ity
n
ews;
an
d
v
i)
o
th
er
s
s
u
ch
as e
d
u
ca
tio
n
an
d
life
s
ty
les.
T
h
e
av
er
ag
e
n
u
m
b
er
o
f
w
o
r
d
s
in
th
e
n
ews
is
2
1
9
.
0
8
wo
r
d
s
an
d
th
e
lo
n
g
est
ar
ticle
h
as
3
,
3
8
8
wo
r
d
s
.
T
h
e
av
er
a
g
e
n
u
m
b
e
r
o
f
wo
r
d
s
in
a
h
ea
d
lin
e
is
3
.
8
3
.
T
h
e
n
u
m
b
er
o
f
n
ews u
s
ed
f
o
r
tr
ain
in
g
i
s
4
0
,
0
0
0
an
d
5
,
0
0
0
f
o
r
test
in
g
.
T
h
is
t
r
ain
-
test
s
p
lit
o
f
9
0
% a
n
d
1
0
%
will p
r
ev
en
t
th
e
o
v
er
f
itti
n
g
is
s
u
e
o
f
th
e
m
o
d
el
tr
ain
in
g
.
4
.
2
.
T
ex
t
cle
a
nin
g
T
h
e
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
b
eg
an
with
p
u
ttin
g
all
th
e
in
p
u
t
tex
t
in
to
lo
wer
ca
s
e.
T
h
en
,
t
h
e
r
eg
u
lar
ex
p
r
ess
io
n
an
d
t
h
e
b
ea
u
tif
u
l
s
o
u
p
p
a
r
s
er
in
Py
th
o
n
wer
e
em
p
lo
y
ed
to
r
em
o
v
e
an
y
em
b
ed
d
ed
HT
ML
tag
s
an
d
Go
o
g
le
T
a
g
s
f
r
o
m
th
e
in
p
u
t n
ews te
x
t to
p
r
o
d
u
ce
a
clea
n
i
n
p
u
t te
x
t
r
ep
r
esen
tatio
n
.
T
o
f
u
r
t
h
er
r
e
d
u
ce
th
e
n
o
is
e
in
th
e
in
p
u
t
tex
t,
p
u
n
ctu
atio
n
m
ar
k
s
an
d
n
o
n
-
in
f
o
r
m
ativ
e
s
p
ec
ial
ch
ar
ac
ter
s
wer
e
r
e
m
o
v
e
d
b
y
u
s
in
g
th
e
r
eg
u
lar
ex
p
r
ess
io
n
.
T
h
e
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
was
f
u
r
th
er
en
h
an
ce
d
with
th
e
r
em
o
v
al
o
f
all
n
o
n
-
tex
tu
al
-
b
ased
ch
ar
ac
ter
s
u
s
in
g
th
e
r
e
g
u
lar
ex
p
r
ess
io
n
s
in
Py
th
o
n
.
4
.
3
.
I
np
ut
pre
-
pro
ce
s
s
ing
B
ased
o
n
th
e
ex
p
er
im
e
n
t,
th
e
L
STM
an
d
Bi
L
STM
s
eq
2
s
eq
m
o
d
els
u
s
ed
wo
r
d
-
lev
el
to
k
e
n
izatio
n
to
in
p
u
t
th
e
tex
t
i
n
to
th
e
em
b
ed
d
i
n
g
lay
e
r
.
T
h
en
,
th
e
p
o
s
t
-
p
ad
d
i
n
g
was
ap
p
lied
to
th
e
to
k
e
n
ized
in
p
u
t
o
f
th
e
m
o
d
els
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Ma
la
y
p
h
o
n
eme
-
b
a
s
ed
s
u
b
w
o
r
d
n
ews
h
ea
d
lin
e
g
en
era
to
r
fo
r
lo
w
-
r
eso
u
r
ce
la
n
g
u
a
g
e
(
Yeo
n
g
Ts
a
n
n
P
h
u
a
)
4971
to
en
s
u
r
e
all
th
e
i
n
p
u
t
h
ad
a
len
g
th
o
f
1
0
0
w
o
r
d
s
.
On
th
e
o
th
er
h
an
d
,
th
e
MSAS
-
tan
h
a
n
d
MSAS
-
R
eL
U
m
o
d
els’
in
p
u
t
wen
t
th
r
o
u
g
h
s
alien
t
s
en
ten
ce
ex
tr
ac
tio
n
u
s
in
g
th
e
lead
in
g
s
en
ten
ce
n
ews
T
ex
tR
an
k
a
lg
o
r
ith
m
.
T
h
en
,
th
e
ex
tr
ac
ted
s
en
ten
ce
s
wen
t
th
r
o
u
g
h
th
e
s
u
b
wo
r
d
to
k
en
izati
o
n
p
r
o
ce
s
s
u
s
in
g
th
e
Ma
lay
p
h
o
n
em
e
s
u
b
wo
r
d
to
k
en
izer
to
b
r
ea
k
ea
ch
in
p
u
t w
o
r
d
i
n
to
s
u
b
wo
r
d
to
k
en
s
ac
c
o
r
d
in
g
to
th
e
to
k
e
n
izer
’
s
s
y
lla
b
le
s
tr
u
ctu
r
es.
Nex
t,
th
e
p
o
s
t
-
p
a
d
d
in
g
was
a
p
p
lied
to
t
h
e
s
u
b
wo
r
d
to
k
e
n
s
to
e
n
s
u
r
e
all
t
h
e
in
p
u
t
h
ad
a
len
g
th
o
f
1
0
0
wo
r
d
s
.
Su
b
s
eq
u
en
tly
,
all
th
e
in
p
u
ts
wer
e
f
ed
in
to
th
e
em
b
ed
d
in
g
la
y
er
o
f
th
e
s
eq
2
s
eq
m
o
d
el
.
T
h
e
em
b
ed
d
in
g
lay
e
r
s
f
o
r
all
th
e
m
o
d
els we
r
e
co
n
f
ig
u
r
ed
with
3
0
0
d
im
en
s
io
n
s
.
4
.
4
.
M
o
del
building
B
ased
o
n
th
e
in
v
esti
g
atio
n
o
f
li
ter
atu
r
e
[
7
]
,
[
4
8
]
,
th
e
L
STM
-
b
ased
s
eq
2
s
eq
m
o
d
els
ar
e
co
m
m
o
n
ly
u
s
e
d
in
tex
t
s
u
m
m
ar
izatio
n
.
Hen
ce
,
two
s
eq
2
s
eq
m
o
d
els:
L
STM
an
d
B
iLST
M
h
av
e
b
ee
n
s
elec
ted
as
b
aselin
e
m
o
d
els
in
th
e
ex
p
er
im
e
n
t.
I
n
th
e
s
eq
2
s
eq
m
o
d
el,
a
n
en
co
d
er
an
d
a
d
e
co
d
er
with
an
atten
tio
n
m
ec
h
a
n
is
m
ar
e
em
p
l
o
y
e
d
.
E
ac
h
o
f
th
e
s
eq
2
s
eq
m
o
d
els
will
co
n
tain
a
s
in
g
le
R
NN
-
b
ased
lay
er
as
en
co
d
er
an
d
a
s
in
g
le
lay
er
o
f
R
NN
-
b
ased
as
d
ec
o
d
er
.
B
o
th
th
e
en
co
d
er
an
d
d
ec
o
d
er
lay
er
s
wer
e
co
n
f
i
g
u
r
ed
with
3
0
0
h
id
d
en
u
n
its
p
er
la
y
er
.
T
o
p
r
ev
en
t
o
v
er
f
itti
n
g
is
s
u
es,
a
d
r
o
p
o
u
t
r
ate
o
f
0
.
4
was
s
et
to
th
e
n
o
n
-
r
ec
u
r
r
e
n
t
c
o
n
n
ec
tio
n
o
f
t
h
e
n
eu
r
o
n
s
d
u
r
in
g
th
e
tr
ain
in
g
.
At
th
e
s
am
e
tim
e,
r
eg
u
lar
izatio
n
is
ad
d
ed
in
th
e
lay
e
r
to
ig
n
o
r
e
s
o
m
e
o
f
th
e
h
id
d
en
s
tate
u
n
its
u
s
in
g
a
0
.
4
r
ec
u
r
r
en
t
d
r
o
p
o
u
t
r
ate.
B
esid
es
th
at,
th
ese
lay
er
s
wer
e
co
n
f
ig
u
r
ed
with
f
u
ll
s
eq
u
en
c
e
o
u
tp
u
ts
an
d
s
tate
r
etu
r
n
s
with
in
th
e
ar
ch
itectu
r
e.
T
h
ese
s
ettin
g
s
will
im
p
r
o
v
e
th
e
r
o
b
u
s
tn
ess
an
d
g
en
er
aliza
tio
n
ab
ilit
y
o
f
th
e
m
o
d
els.
Su
b
s
eq
u
en
tly
,
t
h
e
B
ah
d
an
a
u
et
a
l.
[
7
]
atten
tio
n
m
ec
h
a
n
is
m
was
in
tr
o
d
u
c
ed
to
f
u
r
t
h
er
i
m
p
r
o
v
e
th
e
d
ec
o
d
er
’
s
ca
p
ab
ilit
y
.
I
t
will
allo
w
th
e
d
ec
o
d
er
to
f
o
c
u
s
o
n
th
e
r
elev
an
t
p
ar
ts
o
f
th
e
i
n
p
u
t
s
eq
u
e
n
ce
wh
e
n
g
en
er
atin
g
h
ea
d
lin
es.
T
h
e
atte
n
tio
n
m
ec
h
a
n
is
m
will
co
m
p
u
t
e
th
e
atten
tio
n
s
co
r
e
b
ased
o
n
t
h
e
en
co
d
er
’
s
h
id
d
en
s
tates
ac
co
r
d
in
g
to
th
e
alig
n
m
en
t
to
th
e
d
ec
o
d
er
’
s
cu
r
r
en
t
s
tate.
I
n
th
e
ex
p
er
im
en
t,
th
e
atte
n
tio
n
m
ec
h
an
is
m
’
s
alig
n
m
en
t
s
co
r
e
f
o
r
L
STM
,
B
iLST
M,
an
d
MSAS
-
tan
h
wer
e
co
m
p
u
ted
u
s
in
g
th
e
h
y
p
er
b
o
li
c
tan
g
en
t
f
u
n
ctio
n
.
I
n
th
e
MSAS
-
R
eL
U
m
o
d
el,
th
e
alig
n
m
en
t sco
r
e
was c
o
m
p
u
t
ed
u
s
in
g
th
e
R
eL
U
f
u
n
ctio
n
.
4
.
5
.
T
ra
ini
ng
T
o
en
s
u
r
e
th
e
ef
f
ec
tiv
e
n
ess
o
f
th
e
m
o
d
el
lear
n
i
n
g
an
d
p
r
e
v
en
t
o
v
er
f
itti
n
g
,
all
th
e
m
o
d
e
ls
’
tr
ain
in
g
ad
o
p
ted
th
e
R
MSp
r
o
p
o
p
tim
izer
with
a
d
ef
au
lt
lear
n
in
g
r
ate
o
f
0
.
0
0
0
1
.
T
h
e
‘
s
p
ar
s
e_
ca
teg
o
r
ical_
cr
o
s
s
en
tr
o
p
y
’
lo
s
s
f
u
n
ctio
n
was
ch
o
s
en
as
th
is
f
u
n
ctio
n
co
u
ld
h
an
d
le
th
e
p
r
ed
ictio
n
o
f
th
e
n
e
x
t
to
k
e
n
p
r
o
b
lem
.
T
h
e
m
in
im
iz
e
lo
s
s
g
o
al
was
im
p
lem
en
ted
in
t
o
th
e
t
r
ain
in
g
p
r
o
ce
s
s
an
d
th
e
v
alid
atio
n
lo
s
s
was
m
o
n
ito
r
ed
.
T
h
e
v
alid
atio
n
lo
s
s
is
th
e
in
d
icato
r
o
f
t
h
e
m
o
d
el’
s
g
en
er
aliza
tio
n
ca
p
a
b
ilit
y
.
T
o
p
r
ev
en
t
ex
tr
a
t
r
ain
in
g
e
p
o
ch
s
an
d
r
ed
u
ce
o
v
er
f
itti
n
g
r
is
k
,
ea
r
ly
s
to
p
p
in
g
ca
llb
ac
k
was
in
tr
o
d
u
ce
d
to
th
e
m
o
d
el
tr
ain
in
g
.
All
th
e
m
o
d
el
tr
ai
n
in
g
was
co
n
d
u
cted
o
v
er
a
m
ax
im
u
m
o
f
5
0
e
p
o
ch
s
with
a
b
atch
s
ize
o
f
6
4
to
b
alan
ce
t
h
e
co
m
p
u
tatio
n
al
ef
f
icien
cy
w
ith
m
em
o
r
y
u
s
ag
e.
T
h
e
ex
p
er
im
en
ts
wer
e
p
er
f
o
r
m
ed
u
s
in
g
an
I
n
tel
i
7
PC
with
3
2
GB
o
f
R
AM
,
an
d
a
n
Nv
id
ia
GT
X1
0
8
0
T
i
GPU
ca
r
d
with
1
2
GB
o
f
GPU
R
AM
.
All
th
e
m
o
d
els
we
r
e
d
ev
el
o
p
ed
u
s
in
g
th
e
Py
t
h
o
n
ten
s
o
r
f
lo
w
lib
r
ar
y
r
u
n
n
in
g
o
n
th
e
Ub
u
n
tu
Desk
to
p
o
p
e
r
atin
g
s
y
s
tem
.
4
.
6
.
Resul
t
s
a
nd
dis
cu
s
s
io
n
T
h
e
o
b
jectiv
e
o
f
t
h
e
f
o
llo
win
g
ex
p
er
im
en
ts
is
to
co
m
p
ar
e
th
e
g
en
er
ated
h
ea
d
lin
es
o
f
th
e
p
r
o
p
o
s
ed
n
ew
MSAS
m
o
d
els
ag
ain
s
t
th
e
L
ST
M
an
d
B
iLST
M
s
eq
2
s
eq
m
o
d
e
ls
.
T
h
e
e
x
p
er
im
e
n
tal
r
esu
lts
w
er
e
ev
alu
ate
d
u
s
in
g
th
e
R
OUGE
ev
alu
atio
n
m
etr
ic,
an
d
th
e
r
esu
lts
d
em
o
n
s
tr
ated
a
s
ig
n
if
ican
t
im
p
r
o
v
e
m
en
t
in
th
e
h
ea
d
lin
e
g
en
er
atio
n
p
er
f
o
r
m
a
n
ce
c
o
m
p
ar
ed
to
o
th
e
r
r
e
p
o
r
ted
m
eth
o
d
s
,
B
iLST
M
an
d
L
STM
as
d
el
in
ea
ted
in
T
ab
le
2
.
T
h
e
p
r
o
p
o
s
ed
MSAS
m
o
d
els
h
av
e
r
ep
o
r
ted
th
e
f
o
llo
win
g
ex
p
er
im
e
n
tal
r
esu
lts
:
i
)
th
e
MSAS
-
tan
h
m
o
d
el
ac
h
iev
ed
1
6
.
6
2
R
OUGE
-
1
,
4
.
3
2
R
OUGE
-
2
,
a
n
d
1
6
.
2
3
R
O
UGE
-
L
,
an
d
ii
)
th
e
MSAS
-
R
e
L
U
ac
h
iev
ed
1
8
.
8
1
R
OUGE
-
1
,
5
.
0
7
R
OUGE
-
2
,
an
d
1
8
.
3
5
R
OUGE
-
L
.
Acc
o
r
d
in
g
to
th
e
r
ep
o
r
ted
ex
p
e
r
im
en
tal
r
esu
lts
,
b
o
th
m
o
d
els
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
ed
th
e
s
tan
d
ar
d
L
STM
-
b
ased
m
o
d
el
s
.
No
tab
ly
,
b
o
t
h
MSAS
-
b
ased
m
o
d
els
r
ec
o
r
d
ed
a
m
in
im
u
m
o
f
1
1
.
0
0
im
p
r
o
v
em
en
t in
R
OUGE
-
1
an
d
R
OUGE
-
L
s
co
r
es.
Am
o
n
g
all,
th
e
MSAS
-
R
eL
U
ex
h
ib
ited
th
e
m
o
s
t sig
n
if
ican
t im
p
r
o
v
em
en
t,
as r
ep
o
r
te
d
in
T
a
b
le
2
b
y
i
ts
R
OU
GE
s
co
r
es.
T
ab
le
2
.
R
OUGE
s
co
r
es f
o
r
th
e
m
o
d
els
R
O
U
G
E
-
1
R
O
U
G
E
-
2
R
O
U
G
E
-
L
M
o
d
e
l
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
P
r
e
c
i
s
i
o
n
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
B
i
LST
M
5
.
6
3
4
.
7
3
4
.
9
8
0
.
8
2
0
.
6
9
0
.
7
1
0
.
8
2
5
.
5
7
4
.
6
9
4
.
9
3
LSTM
5
.
1
4
4
.
0
0
4
.
3
9
1
.
0
0
0
.
8
4
0
.
8
9
1
.
0
0
5
.
0
8
3
.
9
5
4
.
3
3
M
S
A
S
-
t
a
n
h
1
8
.
7
3
1
6
.
0
2
1
6
.
6
2
5
.
0
1
4
.
1
2
4
.
3
2
5
.
0
1
1
8
.
2
9
1
5
.
6
5
1
6
.
2
3
M
S
A
S
-
R
e
LU
2
1
.
2
9
1
8
.
0
4
1
8
.
8
1
5
.
8
8
4
.
9
6
5
.
0
7
5
.
8
8
2
0
.
7
9
1
7
.
6
0
1
8
.
3
5
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
9
6
5
-
4
9
7
5
4972
Fu
r
th
er
m
o
r
e
,
T
ab
le
3
s
h
o
ws
th
e
av
er
ag
e
OOV
r
ates
f
o
r
th
e
B
iLST
M,
L
STM
,
M
SAS
-
tan
h
,
an
d
MSAS
-
R
eL
U
te
s
t
m
o
d
els.
B
ased
o
n
th
e
test
r
esu
lts
,
th
e
MSAS
-
b
ased
m
o
d
els
d
e
m
o
n
s
tr
ate
s
ig
n
if
ican
t
im
p
r
o
v
em
e
n
t in
h
ea
d
lin
e
g
en
e
r
atio
n
ca
p
ab
ilit
ies co
m
p
ar
ed
t
o
th
e
o
th
er
two
m
o
d
els.
Acc
o
r
d
in
g
to
T
ab
le
3
,
th
e
MSAS
-
b
ased
m
o
d
els
m
an
ag
ed
to
r
ed
u
ce
th
e
OOV
r
ates
s
i
g
n
if
ican
t
ly
.
MSAS
-
b
ased
m
o
d
els
ac
h
iev
ed
b
elo
w
1
6
%
o
f
th
e
OOV
r
ate
with
MS
AS
-
tan
h
at
1
5
.
5
%
an
d
R
MA
S
-
R
eL
U
at
1
4
.
5
%,
r
esp
ec
tiv
ely
.
T
h
ese
r
esu
lts
s
h
o
w
th
at
MSAS
-
b
ased
m
o
d
els
ar
e
ca
p
ab
le
o
f
r
ed
u
cin
g
th
e
v
o
ca
b
u
lar
y
lim
itatio
n
t
h
at
o
f
ten
o
cc
u
r
s
in
m
o
s
t
o
f
th
e
s
eq
2
s
eq
m
o
d
e
ls
th
at
ar
e
wid
ely
u
s
ed
in
h
ea
d
li
n
e
g
e
n
er
ati
o
n
.
T
h
is
n
o
tab
le
im
p
r
o
v
em
en
t
s
u
g
g
ests
th
at
th
e
p
r
o
p
o
s
ed
m
o
d
el
ca
n
o
v
er
co
m
e
th
e
lar
g
e
v
o
ca
b
u
lar
y
s
ize
r
e
q
u
ir
em
en
t
in
t
h
e
em
b
ed
d
in
g
l
ay
er
s
o
f
th
e
m
o
d
el,
wh
ich
is
h
elp
f
u
l
f
o
r
th
e
Ma
lay
lan
g
u
ag
e
th
at
h
as
co
m
p
le
x
m
o
r
p
h
o
lo
g
ical
s
tr
u
ctu
r
es.
B
esid
es
th
at,
th
e
ap
p
r
o
ac
h
allo
ws a
m
o
r
e
co
m
p
ac
t r
ep
r
es
en
tatio
n
o
f
t
h
e
v
o
ca
b
u
lar
y
.
T
ab
le
3
.
Mo
d
el
an
aly
s
is
f
o
r
t
h
e
m
o
d
els
M
o
d
e
l
O
O
V
r
a
t
e
(
%)
B
i
LST
M
6
0
.
7
LSTM
7
4
.
7
M
S
A
S
-
t
a
n
h
1
5
.
5
M
S
A
S
-
R
e
LU
1
4
.
5
I
t
is
al
s
o
r
ep
o
r
ted
in
T
ab
le
2
,
wh
en
co
m
p
ar
in
g
th
e
two
M
SAS
m
o
d
els,
th
e
R
eL
U
atten
t
io
n
m
o
d
el
d
em
o
n
s
tr
ates
s
ig
n
if
ican
t
im
p
r
o
v
em
en
t
in
t
h
e
n
u
m
er
ical
q
u
a
n
tity
o
f
2
.
1
9
f
o
r
R
OUGE
-
1
,
0
.
7
5
f
o
r
R
OUGE
-
2
,
an
d
2
.
1
2
f
o
r
R
OUGE
-
L
co
m
p
ar
ed
to
MSAS
-
tan
h
.
Ad
d
itio
n
ally
,
th
e
MSAS
-
R
eL
U
m
o
d
el’
s
OOV
r
ate
is
also
r
ed
u
ce
d
b
y
1
.
0
0
%
an
d
ex
h
ib
its
g
r
ea
t
co
m
p
u
tatio
n
al
ef
f
icien
cy
th
at
o
n
ly
r
eq
u
ir
es
6
1
6
.
6
3
s
e
co
n
d
s
p
er
ep
o
ch
f
o
r
tr
ain
in
g
as
co
m
p
a
r
ed
to
th
e
M
SAM
-
tan
h
m
o
d
el,
wh
ich
r
eq
u
ir
es
an
av
er
ag
e
o
f
6
2
4
.
6
s
ec
o
n
d
s
.
T
h
ese
f
in
d
in
g
s
in
d
icat
e
th
e
co
m
p
u
tatio
n
al
ef
f
icien
cy
an
d
ef
f
ec
tiv
en
ess
o
f
th
e
MSAS
-
R
eL
U
m
o
d
el
in
in
f
er
en
ce
an
d
lear
n
in
g
f
r
o
m
d
ata.
Fig
u
r
e
2
s
h
o
ws
t
h
e
ex
am
p
le
o
f
a
s
u
m
m
ar
y
p
r
o
d
u
c
ed
b
y
all
th
e
m
o
d
els
in
t
h
e
e
x
p
er
im
en
t.
Fr
o
m
t
h
e
s
am
p
le,
th
e
o
u
tp
u
t
f
r
o
m
th
e
L
STM
ap
p
ea
r
s
to
h
av
e
th
e
OOV
is
s
u
e.
T
h
e
o
u
tp
u
t
o
f
B
iLST
M
an
d
MSAS
-
tan
h
wer
e
ab
le
to
p
r
o
d
u
ce
r
elate
d
h
ea
d
lin
e
b
u
t
th
e
co
n
te
x
t
is
n
o
t
c
o
r
r
ec
t.
T
h
e
MSAS
-
R
eL
U
m
o
d
el
p
r
o
d
u
ce
d
a
m
o
r
e
f
lu
en
t h
ea
d
lin
e
.
I
1
:
“
ta
n
a
h
mera
h
ma
y
a
t
r
ema
ja
lela
ki
ya
n
g
terj
u
n
ke
s
u
n
g
a
i
j
en
o
b
b
u
kit
b
u
n
g
a
d
i
s
in
i
r
a
b
u
l
a
lu
d
itemu
i
d
u
a
kilo
mete
r
d
a
r
i
lo
ka
s
i
ma
n
g
s
a
d
ila
p
o
r
ka
n
h
ila
n
g
ke
lma
r
in
ma
n
g
s
a
ya
n
g
b
eru
s
ia
ta
h
u
n
d
itemu
i
a
n
g
g
o
ta
men
ca
r
i
d
a
n
men
ye
la
ma
t
ja
m
p
eta
n
g
ke
tu
a
p
o
lis
d
a
era
h
ta
n
a
h
mera
h
d
e
p
u
ti
s
u
p
erin
ten
d
a
n
s
u
z
a
imi
mo
h
a
ma
d
b
erka
ta
ma
ya
t
ma
n
g
s
a
d
itemu
i
ters
a
n
g
k
u
t
p
a
d
a
k
a
yu
d
i
s
u
n
g
a
i
b
erken
a
a
n
d
a
n
ma
s
ih
b
erp
a
ka
ia
n
len
g
ka
p
…”
[
49
]
G:
ma
ya
t rema
ja
ters
a
n
g
ku
t k
a
yu
B
iL
ST
M
:
ma
ya
t m
a
ya
t d
a
la
m
b
ilik ta
n
d
a
s
L
S
T
M
:
u
p
d
ate
OOV
OOV
O
OV
M
SA
S
-
t
a
nh
:
ma
ya
t rema
ja
terj
u
n
s
u
n
g
a
i
M
SA
S
-
ReLU:
ma
ya
t rema
ja
d
itemu
i
I
2
:
“
s
e
r
a
m
a
i
a
h
l
i
s
i
n
d
i
k
e
t
m
a
c
a
u
s
c
a
m
d
i
t
a
h
a
n
s
e
j
a
k
a
w
a
l
t
a
h
u
n
i
n
i
m
e
m
b
a
b
i
t
k
a
n
k
e
r
u
g
i
a
n
r
m
j
u
t
a
s
e
t
a
k
a
t
n
o
v
e
m
b
e
r
s
e
l
e
p
a
s
m
e
m
p
e
r
d
a
y
a
m
a
n
g
s
a
d
i
k
e
d
a
h
k
e
t
u
a
j
a
b
a
t
a
n
s
i
a
s
a
t
a
n
j
e
n
a
y
a
h
k
o
m
e
r
s
i
l
k
e
d
a
h
s
u
p
e
r
i
n
t
e
n
d
a
n
c
h
a
n
t
e
c
k
p
a
i
n
g
b
e
r
k
a
t
a
a
n
g
k
a
i
t
u
j
u
g
a
m
e
n
u
n
j
u
k
k
a
n
p
e
n
i
n
g
k
a
t
a
n
b
e
r
b
a
n
d
i
n
g
t
a
h
u
n
l
a
l
u
…
”
[5
0
]
G:
r
a
ma
i o
r
a
n
g
ke
d
a
h
ke
n
a
tip
u
ma
ca
u
s
ca
m
B
iL
ST
M
:
p
o
lis
ca
r
i le
la
ki
w
a
r
g
a
a
s
in
g
d
ita
h
a
n
L
S
T
M
:
OOV
O
OV
OOV
M
SA
S
-
t
a
nh
:
s
in
d
iket
ma
ca
u
s
ca
m
d
ita
h
a
n
M
SA
S
-
ReLU:
s
in
d
iket
ma
ca
u
s
ca
m
d
icek
u
p
Fig
u
r
e
2
.
E
x
am
p
le
h
ea
d
lin
e
p
r
o
d
u
ce
d
.
I
1
a
n
d
I
2
ar
e
t
h
e
in
p
u
t,
an
d
G
is
th
e
ac
tu
al
h
ea
d
lin
e
B
ased
o
n
th
e
f
in
d
in
g
s
,
th
e
M
SAS
-
R
eL
U
m
o
d
el
d
em
o
n
s
tr
ated
s
u
p
er
io
r
p
e
r
f
o
r
m
an
ce
,
wh
i
ch
ca
n
b
e
attr
ib
u
ted
to
s
ev
er
al
f
ac
to
r
s
.
A
d
o
p
tin
g
th
e
R
eL
U
atten
tio
n
s
c
o
r
e
m
ec
h
an
is
m
lik
ely
h
el
p
s
allev
iate
th
e
v
an
is
h
in
g
g
r
ad
ien
t
p
r
o
b
lem
a
n
d
m
ain
tai
n
s
b
etter
g
r
ad
ien
t
f
lo
w
i
n
t
h
e
b
ac
k
p
r
o
p
a
g
atio
n
in
t
h
e
m
o
d
e
l.
B
esid
es
th
at,
th
e
R
eL
U
f
u
n
ctio
n
is
m
o
r
e
co
m
p
u
tatio
n
ally
ef
f
icien
t
co
m
p
ar
ed
to
th
e
T
an
h
f
u
n
ctio
n
d
u
e
to
its
s
im
p
ler
m
ath
em
atica
l
o
p
er
atio
n
s
th
at
d
ir
ec
tly
tr
an
s
late
in
to
th
e
s
am
e
ex
p
ec
tatio
n
in
co
m
p
u
tatio
n
.
Ov
er
all,
we
s
ee
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Ma
la
y
p
h
o
n
eme
-
b
a
s
ed
s
u
b
w
o
r
d
n
ews
h
ea
d
lin
e
g
en
era
to
r
fo
r
lo
w
-
r
eso
u
r
ce
la
n
g
u
a
g
e
(
Yeo
n
g
Ts
a
n
n
P
h
u
a
)
4973
r
ed
u
ctio
n
i
n
OOV
r
ates,
wh
ich
ca
n
b
e
r
elate
d
to
t
h
e
im
p
le
m
en
tatio
n
o
f
th
e
p
h
o
n
e
m
e
-
b
a
s
ed
cu
s
to
m
s
u
b
wo
r
d
em
b
ed
d
in
g
.
T
h
is
ap
p
r
o
ac
h
ef
f
ec
tiv
ely
ad
d
r
ess
es
th
e
v
o
ca
b
u
l
ar
y
lim
itatio
n
th
r
o
u
g
h
th
e
d
ec
o
m
p
o
s
itio
n
o
f
w
o
r
d
s
in
to
m
an
ag
ea
b
le
s
u
b
wo
r
d
s
.
5.
CO
NCLU
SI
O
N
T
h
is
p
ap
er
p
r
esen
ts
a
co
m
p
ar
is
o
n
b
etwe
en
co
n
tem
p
o
r
a
r
y
R
NN
-
b
ased
m
o
d
els
an
d
th
e
n
ewly
p
r
o
p
o
s
ed
MSAS
m
o
d
els
to
ad
d
r
ess
th
e
g
en
er
ati
o
n
o
f
Ma
lay
lan
g
u
a
g
e
n
ews
h
ea
d
lin
es.
Du
e
to
th
e
lim
itatio
n
o
f
th
e
ex
is
ten
ce
o
f
a
n
o
p
en
Ma
la
y
la
n
g
u
ag
e
n
ews
d
ataset
av
ailab
le
f
o
r
th
e
e
x
p
er
im
e
n
t,
th
e
two
R
NN
-
b
ased
m
o
d
els
wer
e
s
et
to
th
e
s
am
e
p
r
o
p
er
t
y
d
ataset
th
at
was
co
llected
s
p
e
cif
ically
f
o
r
th
is
r
esear
ch
.
B
ased
o
n
th
e
em
p
ir
ical
ev
alu
atio
n
o
f
th
e
e
x
p
er
im
e
n
t
r
esu
lt,
th
e
n
ewly
p
r
o
p
o
s
e
d
m
o
d
el,
MSAS
-
R
eL
U,
d
e
m
o
n
s
tr
ated
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
o
v
e
r
th
e
B
iLST
M,
L
STM
,
an
d
MSAS
-
tan
h
m
o
d
els in
th
e
ex
p
er
im
en
ts
.
T
h
e
MSAS
-
R
eL
U
m
o
d
el
s
h
o
ws
its
alg
o
r
ith
m
ic
lea
r
n
in
g
an
d
in
f
er
en
ce
ef
f
ec
tiv
e
n
ess
,
p
a
r
ticu
lar
ly
in
th
e
ar
ea
s
o
f
im
p
r
o
v
em
en
t
in
R
OUGE
s
co
r
es
an
d
r
ed
u
ctio
n
o
f
th
e
O
OV
r
ate.
T
h
ese
n
o
tab
le
ex
p
er
i
m
en
tal
ac
h
iev
em
en
ts
clea
r
ly
h
ig
h
lig
h
t
th
e
p
r
o
p
o
s
ed
n
ew
MSAS
h
y
b
r
id
ex
tr
ac
t
iv
e
an
d
ab
s
tr
ac
tiv
e
s
u
m
m
ar
izatio
n
m
o
d
els
im
p
lem
en
ted
to
g
eth
er
with
p
h
o
n
e
m
e
-
b
ased
s
u
b
wo
r
d
em
b
ed
d
in
g
s
ar
e
ab
le
to
g
en
er
ate
m
o
r
e
ac
c
u
r
ate
n
ews
h
ea
d
lin
e
s
ev
en
th
o
u
g
h
th
e
p
r
o
p
o
s
ed
MSAS
m
o
d
els
ar
e
lear
n
in
g
an
d
in
f
e
r
en
ce
in
d
at
asets
with
u
n
b
alan
ce
d
co
n
s
tr
ain
ts
o
f
v
o
ca
b
u
lar
y
lim
itatio
n
.
I
n
s
h
o
r
t,
th
e
p
r
o
p
o
s
ed
n
ew
MSAS
m
o
d
els
h
av
e
s
ig
n
if
ican
tly
s
o
lv
ed
th
e
co
m
p
lex
m
o
r
p
h
o
lo
g
ical
s
tr
u
ctu
r
e
an
d
lo
w
-
r
eso
u
r
ce
tr
a
in
in
g
is
s
u
e
co
m
m
o
n
ly
f
o
u
n
d
i
n
Ma
lay
lan
g
u
ag
e
-
b
ased
co
m
p
u
tatio
n
al
lin
g
u
is
tic
ap
p
licatio
n
s
.
T
h
e
MSAS
m
o
d
el
’
s
ex
p
er
im
en
tal
s
u
cc
ess
c
lear
ly
in
d
icate
s
th
at
it
i
s
a
g
u
id
elin
e
f
o
r
o
th
er
r
esear
ch
er
s
to
im
p
le
m
en
t
f
u
tu
r
e
Ma
lay
lan
g
u
ag
e
NHG
alg
o
r
i
th
m
s
u
s
in
g
MSAS
a
s
th
e
f
o
u
n
d
atio
n
m
o
d
el.
Als
o
,
th
e
ac
h
iev
em
en
t
o
f
t
h
e
MSAS
r
o
le
m
o
d
el
s
o
lid
if
ies
a
n
ew
b
en
ch
m
ar
k
f
o
r
Ma
lay
MH
G
task
s
.
T
h
is
r
esear
ch
p
av
es
th
e
way
f
o
r
f
u
tu
r
e
Ma
l
ay
lan
g
u
a
g
e
NHG
ex
p
lo
r
atio
n
to
in
v
e
n
t
o
th
er
a
d
v
an
ce
d
a
n
d
m
o
d
er
n
s
eq
2
s
eq
m
o
d
els,
s
p
ec
if
ically
in
th
e
lin
g
u
is
tic
ap
p
licatio
n
u
s
in
g
t
h
e
Ma
lay
lan
g
u
ag
e
.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
is
r
esear
ch
was
f
u
lly
f
u
n
d
ed
b
y
th
e
Yay
asan
UT
P
-
PR
G
C
y
cle
2
0
2
2
(
0
1
5
PB
C
-
0
0
5
)
an
d
D
ep
ar
tm
en
t
o
f
C
o
m
p
u
ter
a
n
d
I
n
f
o
r
m
atio
n
Scien
ce
at
Un
iv
er
s
iti
T
ek
n
o
lo
g
i
PET
R
ONAS.
T
h
e
au
th
o
r
s
wo
u
ld
lik
e
to
th
an
k
Yay
asan
UT
P
-
PR
G
f
o
r
s
u
p
p
o
r
tin
g
th
is
wo
r
k
.
RE
F
E
R
E
NC
E
S
[
1
]
G
.
M
.
M
o
n
t
e
j
o
a
n
d
T.
Q
.
A
d
r
i
a
n
o
,
“
A
c
r
i
t
i
c
a
l
d
i
s
c
o
u
r
se
a
n
a
l
y
si
s
o
f
h
e
a
d
l
i
n
e
s
i
n
o
n
l
i
n
e
n
e
w
s
p
o
r
t
a
l
s,
”
J
o
u
rn
a
l
o
f
Ad
v
a
n
c
e
s
i
n
H
u
m
a
n
i
t
i
e
s
a
n
d
S
o
c
i
a
l
S
c
i
e
n
c
e
s
,
v
o
l
.
4
,
n
o
.
2
,
A
p
r
.
2
0
1
8
,
d
o
i
:
1
0
.
2
0
4
7
4
/
j
a
h
ss
-
4
.
2
.
2
.
[
2
]
K
.
K
a
i
k
h
a
h
,
“
A
u
t
o
m
a
t
i
c
t
e
x
t
s
u
mm
a
r
i
z
a
t
i
o
n
w
i
t
h
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
i
n
2
0
0
4
2
n
d
I
n
t
e
r
n
a
t
i
o
n
a
l
I
E
EE
C
o
n
f
e
r
e
n
c
e
o
n
'
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
s'
,
I
EEE,
2
0
0
4
,
p
p
.
4
0
–
4
4
.
d
o
i
:
1
0
.
1
1
0
9
/
I
S
.
2
0
0
4
.
1
3
4
4
6
3
4
.
[
3
]
O
.
Ti
l
k
a
n
d
T
.
A
l
u
m
ä
e
,
“
L
o
w
-
r
e
s
o
u
r
c
e
n
e
u
r
a
l
h
e
a
d
l
i
n
e
g
e
n
e
r
a
t
i
o
n
,
”
a
rX
i
v
-
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
p
p
.
1
-
7
,
Ju
l
.
2
0
1
7
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
7
0
7
.
0
9
7
6
9
.
[
4
]
R
.
Z
h
a
n
g
,
J.
G
u
o
,
Y
.
F
a
n
,
Y
.
La
n
,
a
n
d
X
.
C
h
e
n
g
,
“
S
t
r
u
c
t
u
r
e
l
e
a
r
n
i
n
g
f
o
r
h
e
a
d
l
i
n
e
g
e
n
e
r
a
t
i
o
n
,
”
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
AAAI
C
o
n
f
e
r
e
n
c
e
o
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
3
4
,
n
o
.
5
,
p
p
.
9
5
5
5
–
9
5
6
2
,
A
p
r
.
2
0
2
0
,
d
o
i
:
1
0
.
1
6
0
9
/
a
a
a
i
.
v
3
4
i
0
5
.
6
5
0
1
.
[
5
]
B
.
K
.
B
o
g
u
r
a
e
v
a
n
d
M
.
S
.
N
e
f
f
,
“
D
i
s
c
o
u
r
se
se
g
me
n
t
a
t
i
o
n
i
n
a
i
d
o
f
d
o
c
u
me
n
t
su
m
mari
z
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
3
3
r
d
A
n
n
u
a
l
H
a
w
a
i
i
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
S
y
st
e
m
S
c
i
e
n
c
e
s
,
I
EEE
C
o
m
p
u
t
.
S
o
c
,
2
0
0
0
,
d
o
i
:
1
0
.
1
1
0
9
/
H
I
C
S
S
.
2
0
0
0
.
9
2
6
6
8
7
.
[
6
]
Z.
Li
,
J.
W
u
,
J.
M
i
a
o
,
a
n
d
X
.
Y
u
,
“
N
e
w
s
h
e
a
d
l
i
n
e
g
e
n
e
r
a
t
i
o
n
b
a
se
d
o
n
i
m
p
r
o
v
e
d
d
e
c
o
d
e
r
f
r
o
m
t
r
a
n
sf
o
r
m
e
r
,
”
S
c
i
e
n
t
i
f
i
c
R
e
p
o
rt
s
,
v
o
l
.
1
2
,
n
o
.
1
,
J
u
l
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
022
-
1
5
8
1
7
-
z.
[
7
]
D
.
B
a
h
d
a
n
a
u
,
K
.
C
h
o
,
a
n
d
Y
.
B
e
n
g
i
o
,
“
N
e
u
r
a
l
m
a
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
b
y
j
o
i
n
t
l
y
l
e
a
r
n
i
n
g
t
o
a
l
i
g
n
a
n
d
t
r
a
n
s
l
a
t
e
,
”
a
rX
i
v
-
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
p
p
.
1
-
1
5
,
S
e
p
.
2
0
1
4
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
4
0
9
.
0
4
7
3
.
[
8
]
E.
M
.
I
z
h
i
k
e
v
i
c
h
,
“
S
i
m
p
l
e
m
o
d
e
l
o
f
s
p
i
k
i
n
g
n
e
u
r
o
n
s,
”
I
EE
E
T
r
a
n
sa
c
t
i
o
n
s
o
n
N
e
u
ra
l
N
e
t
w
o
rks
,
v
o
l
.
1
4
,
n
o
.
6
,
p
p
.
1
5
6
9
–
1
5
7
2
,
N
o
v
.
2
0
0
3
,
d
o
i
:
1
0
.
1
1
0
9
/
TN
N
.
2
0
0
3
.
8
2
0
4
4
0
.
[
9
]
R
.
N
a
l
l
a
p
a
t
i
,
B
.
Zh
o
u
,
C
.
D.
S
a
n
t
o
s,
C
.
G
u
l
c
e
h
r
e
,
a
n
d
B
.
X
i
a
n
g
,
“
A
b
st
r
a
c
t
i
v
e
t
e
x
t
s
u
mm
a
r
i
z
a
t
i
o
n
u
si
n
g
se
q
u
e
n
c
e
-
t
o
-
seq
u
e
n
c
e
R
N
N
s
a
n
d
b
e
y
o
n
d
,
”
i
n
T
h
e
2
0
t
h
S
I
G
N
L
L
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
N
a
t
u
ra
l
L
a
n
g
u
a
g
e
L
e
a
r
n
i
n
g
,
S
t
r
o
u
d
sb
u
r
g
,
P
A
,
U
S
A
:
A
sso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
Li
n
g
u
i
st
i
c
s
,
2
0
1
6
,
p
p
.
2
8
0
–
290
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
K
1
6
-
1
0
2
8
.
[
1
0
]
M
.
B
a
n
k
o
,
V
.
O
.
M
i
t
t
a
l
,
a
n
d
M
.
J
.
W
i
t
b
r
o
c
k
,
“
H
e
a
d
l
i
n
e
g
e
n
e
r
a
t
i
o
n
b
a
se
d
o
n
st
a
t
i
st
i
c
a
l
t
r
a
n
s
l
a
t
i
o
n
,
”
i
n
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
3
8
t
h
An
n
u
a
l
Me
e
t
i
n
g
o
n
Asso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
-
A
C
L
’
0
0
,
M
o
r
r
i
st
o
w
n
,
N
J
,
U
S
A
:
A
sso
c
i
a
t
i
o
n
f
o
r
C
o
mp
u
t
a
t
i
o
n
a
l
Li
n
g
u
i
s
t
i
c
s,
2
0
0
0
,
p
p
.
3
1
8
–
3
2
5
,
d
o
i
:
1
0
.
3
1
1
5
/
1
0
7
5
2
1
8
.
1
0
7
5
2
5
9
.
[
1
1
]
D
.
Za
j
i
c
,
B
.
D
o
r
r
,
a
n
d
R
.
S
c
h
w
a
r
t
z
,
“
A
u
t
o
m
a
t
i
c
h
e
a
d
l
i
n
e
g
e
n
e
r
a
t
i
o
n
f
o
r
n
e
w
s
p
a
p
e
r
st
o
r
i
e
s,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
AC
L
-
2
0
0
2
Wo
r
k
sh
o
p
o
n
T
e
x
t
S
u
m
m
a
ri
z
a
t
i
o
n
,
2
0
0
2
.
[
1
2
]
A
.
K
a
r
a
k
a
n
t
a
,
J.
D
e
h
d
a
r
i
,
a
n
d
J
.
V.
G
e
n
a
b
i
t
h
,
“
N
e
u
r
a
l
m
a
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
f
o
r
l
o
w
-
r
e
s
o
u
r
c
e
l
a
n
g
u
a
g
e
s
w
i
t
h
o
u
t
p
a
r
a
l
l
e
l
c
o
r
p
o
r
a
,
”
Ma
c
h
i
n
e
T
r
a
n
s
l
a
t
i
o
n
,
v
o
l
.
3
2
,
n
o
.
1
–
2
,
p
p
.
1
6
7
–
1
8
9
,
J
u
n
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
5
9
0
-
0
1
7
-
9
2
0
3
-
5.
[
1
3
]
U
.
K
u
mar,
V
.
S
i
n
g
h
,
C
.
A
n
d
r
e
w
,
S
.
R
e
d
d
y
,
a
n
d
A
.
D
a
s,
“
C
o
n
s
o
n
a
n
t
-
v
o
w
e
l
seq
u
e
n
c
e
s
a
s
s
u
b
w
o
r
d
u
n
i
t
s
f
o
r
c
o
d
e
-
mi
x
e
d
l
a
n
g
u
a
g
e
s,”
i
n
32
nd
AAAI
C
o
n
f
e
r
e
n
c
e
o
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
A
AAI
2
0
1
8
,
2
0
1
8
,
p
p
.
8
1
0
3
–
8
1
0
4
,
d
o
i
:
1
0
.
1
6
0
9
/
a
a
a
i
.
v
3
2
i
1
.
1
2
1
9
3
.
[
1
4
]
O
.
K
w
o
n
,
D
.
K
i
m,
S
.
R
.
L
e
e
,
J
.
C
h
o
i
,
a
n
d
S
.
K
.
L
e
e
,
“
H
a
n
d
l
i
n
g
o
u
t
-
of
-
v
o
c
a
b
u
l
a
r
y
p
r
o
b
l
e
m
i
n
h
a
n
g
e
u
l
w
o
r
d
e
m
b
e
d
d
i
n
g
s
,
”
i
n
EAC
L
2
0
2
1
-
1
6
t
h
C
o
n
f
e
re
n
c
e
o
f
t
h
e
Eu
r
o
p
e
a
n
C
h
a
p
t
e
r
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
s
t
i
c
s
,
Pr
o
c
e
e
d
i
n
g
s o
f
t
h
e
C
o
n
f
e
re
n
c
e
,
2
0
2
1
,
p
p
.
3
2
1
3
–
3
2
2
1
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
1
.
e
a
c
l
-
ma
i
n
.
2
8
0
.
[
1
5
]
A
y
a
n
a
e
t
a
l
.
,
“
R
e
c
e
n
t
a
d
v
a
n
c
e
s
o
n
n
e
u
r
a
l
h
e
a
d
l
i
n
e
g
e
n
e
r
a
t
i
o
n
,
”
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
3
2
,
n
o
.
4
,
p
p
.
7
6
8
–
7
8
4
,
J
u
l
.
2
0
1
7
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
3
9
0
-
017
-
1
7
5
8
-
3.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
9
6
5
-
4
9
7
5
4974
[
1
6
]
B
.
D
o
r
r
,
D
.
Za
j
i
c
,
a
n
d
R
.
S
c
h
w
a
r
t
z
,
“
H
e
d
g
e
t
r
i
mm
e
r
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
H
L
T
-
N
AA
C
L
0
3
o
n
T
e
x
t
s
u
m
m
a
r
i
za
t
i
o
n
w
o
rks
h
o
p
,
M
o
r
r
i
st
o
w
n
,
N
J,
U
S
A
:
A
sso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
s
t
i
c
s,
2
0
0
3
,
p
p
.
1
–
8
,
d
o
i
:
1
0
.
3
1
1
5
/
1
1
1
9
4
6
7
.
1
1
1
9
4
6
8
.
[
1
7
]
A
.
K
.
G
a
t
t
a
n
i
,
“
A
u
t
o
ma
t
e
d
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
h
e
a
d
l
i
n
e
g
e
n
e
r
a
t
i
o
n
u
s
i
n
g
d
i
s
c
r
i
mi
n
a
t
i
v
e
mac
h
i
n
e
l
e
a
r
n
i
n
g
m
o
d
e
l
s
,
”
M
.
S
c
.
T
h
e
s
i
s
,
S
c
h
o
o
l
o
f
C
o
m
p
u
t
i
n
g
S
c
i
e
n
c
e
,
S
i
m
o
n
F
r
a
ser U
n
i
v
e
r
si
t
y
,
B
u
r
n
a
b
y
,
C
a
n
a
d
a
,
2
0
1
7
.
[
1
8
]
M
.
H
a
sse
l
a
n
d
N
.
M
a
z
d
a
k
,
“
F
a
r
s
i
S
u
m,”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
Wo
r
k
sh
o
p
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
A
p
p
ro
a
c
h
e
s
t
o
Ar
a
b
i
c
S
c
ri
p
t
-
b
a
s
e
d
L
a
n
g
u
a
g
e
s
-
S
e
m
i
t
i
c
’
0
4
,
M
o
r
r
i
st
o
w
n
,
N
J,
U
S
A
:
A
ss
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
s
t
i
c
s,
2
0
0
4
,
d
o
i
:
1
0
.
3
1
1
5
/
1
6
2
1
8
0
4
.
1
6
2
1
8
2
6
.
[
1
9
]
B
.
D
o
r
r
,
D
.
Za
j
i
c
,
a
n
d
R
.
S
c
h
w
a
r
t
z
,
“
C
r
o
ss
-
l
a
n
g
u
a
g
e
h
e
a
d
l
i
n
e
g
e
n
e
r
a
t
i
o
n
f
o
r
H
i
n
d
i
,
”
AC
M
T
r
a
n
sa
c
t
i
o
n
s
o
n
Asi
a
n
L
a
n
g
u
a
g
e
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
,
v
o
l
.
2
,
n
o
.
3
,
p
p
.
2
7
0
–
2
8
9
,
S
e
p
.
2
0
0
3
,
d
o
i
:
1
0
.
1
1
4
5
/
9
7
9
8
7
2
.
9
7
9
8
7
8
.
[
2
0
]
C
.
A
.
C
o
l
me
n
a
r
e
s,
M
.
Li
t
v
a
k
,
A
.
M
a
n
t
r
a
c
h
,
a
n
d
F
.
S
i
l
v
e
st
r
i
,
“
H
EA
D
S
:
h
e
a
d
l
i
n
e
g
e
n
e
r
a
t
i
o
n
a
s
se
q
u
e
n
c
e
p
r
e
d
i
c
t
i
o
n
u
si
n
g
a
n
a
b
st
r
a
c
t
f
e
a
t
u
r
e
-
r
i
c
h
sp
a
c
e
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
1
5
C
o
n
f
e
re
n
c
e
o
f
t
h
e
N
o
rt
h
A
m
e
ri
c
a
n
C
h
a
p
t
e
r
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s:
H
u
m
a
n
L
a
n
g
u
a
g
e
T
e
c
h
n
o
l
o
g
i
e
s
,
S
t
r
o
u
d
s
b
u
r
g
,
P
A
,
U
S
A
:
A
ss
o
c
i
a
t
i
o
n
f
o
r
C
o
mp
u
t
a
t
i
o
n
a
l
Li
n
g
u
i
st
i
c
s
,
2
0
1
5
,
p
p
.
1
3
3
–
1
4
2
,
d
o
i
:
1
0
.
3
1
1
5
/
v
1
/
N
1
5
-
1
0
1
4
.
[
2
1
]
I
.
M
a
n
i
,
Au
t
o
m
a
t
i
c
S
u
m
m
a
r
i
z
a
t
i
o
n
,
A
mste
r
d
a
m
:
J
o
h
n
B
e
n
j
a
mi
n
s P
u
b
l
i
sh
i
n
g
C
o
m
p
a
n
y
,
2
0
0
1
,
d
o
i
:
1
0
.
1
0
7
5
/
n
l
p
.
3
.
[
2
2
]
S
.
C
h
o
p
r
a
,
M
.
A
u
l
i
,
a
n
d
A
.
M
.
R
u
s
h
,
“
A
b
st
r
a
c
t
i
v
e
se
n
t
e
n
c
e
s
u
mm
a
r
i
z
a
t
i
o
n
w
i
t
h
a
t
t
e
n
t
i
v
e
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
1
6
C
o
n
f
e
re
n
c
e
o
f
t
h
e
N
o
rt
h
Am
e
ri
c
a
n
C
h
a
p
t
e
r
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s:
H
u
m
a
n
L
a
n
g
u
a
g
e
T
e
c
h
n
o
l
o
g
i
e
s
,
S
t
r
o
u
d
s
b
u
r
g
,
P
A
,
U
S
A
:
A
sso
c
i
a
t
i
o
n
f
o
r
C
o
mp
u
t
a
t
i
o
n
a
l
Li
n
g
u
i
st
i
c
s,
2
0
1
6
,
p
p
.
9
3
–
98
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
N
1
6
-
1
0
1
2
.
[
2
3
]
N
.
S
.
K
a
r
i
m,
F
.
M
.
O
n
n
,
H
.
H
.
M
u
sa
,
a
n
d
A
.
H
.
M
a
h
m
o
o
d
,
H
a
l
l
g
r
a
m
m
a
r
t
h
i
r
d
e
d
i
t
i
o
n
(
i
n
M
a
l
a
y
:
T
a
t
a
b
a
h
a
s
a
d
e
w
a
n
e
d
i
si
k
e
t
i
g
a
)
,
K
u
a
l
a
L
u
m
p
u
r
,
M
a
l
a
y
si
a
:
D
e
w
a
n
B
a
h
a
sa
D
a
n
P
u
s
t
a
k
a
,
2
0
1
5
.
[
2
4
]
N
.
M
.
A
w
a
l
,
K
.
A
.
B
a
k
a
r
,
N
.
Z.
A
.
H
a
mi
d
,
a
n
d
N
.
H
.
Ja
l
a
l
u
d
d
i
n
,
“
M
o
r
p
h
o
l
o
g
i
c
a
l
d
i
f
f
e
r
e
n
c
e
s b
e
t
w
e
e
n
b
a
h
a
s
a
m
e
l
a
y
u
a
n
d
e
n
g
l
i
s
h
:
c
o
n
st
r
a
i
n
t
s
i
n
st
u
d
e
n
t
s’
u
n
d
e
r
st
a
n
d
i
n
g
,
”
S
c
h
o
o
l
o
f
L
a
n
g
u
a
g
e
S
t
u
d
i
e
s
& L
i
n
g
u
i
s
t
i
c
s
,
U
n
i
v
e
rs
i
t
i
U
t
a
r
a
M
a
l
a
y
si
a
,
p
p
.
1
-
1
1
,
2
0
0
7
.
[
2
5
]
S
.
A
.
M
.
N
o
a
h
,
N
.
M
.
A
l
i
,
a
n
d
M
.
S
.
H
a
sa
n
,
“
D
e
t
e
r
m
i
n
i
n
g
f
e
a
t
u
r
e
s
o
f
n
e
w
s
h
e
a
d
l
i
n
e
i
n
M
a
l
a
y
n
e
w
s
d
o
c
u
m
e
n
t
(
i
n
M
a
l
a
y
:
P
e
n
e
n
t
u
a
n
f
i
t
u
r
b
a
g
i
p
e
n
g
e
k
st
r
a
k
a
n
t
a
j
u
k
b
e
r
i
t
a
a
k
h
b
a
r
b
a
h
a
s
a
m
e
l
a
y
u
)
,
”
J
o
u
r
n
a
l
o
f
L
a
n
g
u
a
g
e
S
t
u
d
i
e
s
,
v
o
l
.
1
8
,
n
o
.
2
,
p
p
.
1
5
4
–
1
6
7
,
2
0
1
8
,
d
o
i
:
1
0
.
1
7
5
7
6
/
g
e
ma
-
2
0
1
8
-
1
8
0
2
-
1
1
.
[
2
6
]
S
.
A
l
i
a
s
,
S
.
K
.
M
o
h
a
mm
a
d
,
G
.
K
.
H
o
o
n
,
a
n
d
M
.
S
.
S
a
i
n
i
n
,
“
U
n
d
e
r
st
a
n
d
i
n
g
h
u
m
a
n
s
e
n
t
e
n
c
e
c
o
mp
r
e
ssi
o
n
p
a
t
t
e
r
n
f
o
r
M
a
l
a
y
t
e
x
t
su
mm
a
r
i
z
e
r
,
”
i
n
2
0
1
8
F
o
u
rt
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
f
o
rm
a
t
i
o
n
R
e
t
r
i
e
v
a
l
a
n
d
K
n
o
w
l
e
d
g
e
Ma
n
a
g
e
m
e
n
t
(
C
A
MP
)
,
I
EEE,
M
a
r
.
2
0
1
8
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
N
F
R
K
M
.
2
0
1
8
.
8
4
6
4
7
8
8
.
[
2
7
]
M
.
S
.
H
a
sa
n
,
S
.
A
.
M
.
N
o
a
h
,
a
n
d
N
.
M
.
A
l
i
,
“
M
a
l
a
y
t
e
x
t
f
e
a
t
u
r
e
s
f
o
r
a
u
t
o
mat
i
c
n
e
w
s
h
e
a
d
l
i
n
e
g
e
n
e
r
a
t
i
o
n
,
”
J
o
u
rn
a
l
o
f
T
h
e
o
re
t
i
c
a
l
a
n
d
Ap
p
l
i
e
d
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
7
6
,
n
o
.
1
,
p
p
.
3
6
–
4
1
,
2
0
1
5
.
[
2
8
]
R
.
S
i
d
d
a
l
i
n
g
a
p
p
a
a
n
d
K
.
S
e
k
a
r
,
“
B
i
-
d
i
r
e
c
t
i
o
n
a
l
l
o
n
g
s
h
o
r
t
t
e
r
m
m
e
m
o
r
y
u
si
n
g
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
f
o
r
b
i
o
l
o
g
i
c
a
l
e
n
t
i
t
y
r
e
c
o
g
n
i
t
i
o
n
,
”
I
AE
S
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
I
J
-
AI
)
,
v
o
l
.
1
1
,
n
o
.
1
,
p
p
.
8
9
–
1
0
1
,
M
a
r
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
1
.
i
1
.
p
p
8
9
-
1
0
1
.
[
2
9
]
S
.
H
o
c
h
r
e
i
t
e
r
a
n
d
J
.
S
c
h
m
i
d
h
u
b
e
r
,
“
L
o
n
g
sh
o
r
t
-
t
e
r
m m
e
m
o
r
y
,
”
N
e
u
r
a
l
C
o
m
p
u
t
a
t
i
o
n
,
v
o
l
.
9
,
n
o
.
8
,
p
p
.
1
7
3
5
–
1
7
8
0
,
N
o
v
.
1
9
9
7
,
d
o
i
:
1
0
.
1
1
6
2
/
n
e
c
o
.
1
9
9
7
.
9
.
8
.
1
7
3
5
.
[
3
0
]
E
.
M
a
t
h
e
w
a
n
d
S
.
A
b
d
u
l
l
a
,
“
T
h
e
L
S
T
M
t
e
c
h
n
i
q
u
e
f
o
r
d
e
m
a
n
d
f
o
r
e
c
a
s
t
i
n
g
o
f
e
-
p
r
o
c
u
r
e
m
e
n
t
i
n
t
h
e
h
o
s
p
i
t
a
l
i
t
y
i
n
d
u
s
t
r
y
i
n
t
h
e
U
A
E
,
”
I
A
E
S
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
I
J
-
A
I
)
,
v
o
l
.
9
,
n
o
.
4
,
p
p
.
7
5
7
–
7
6
5
,
D
e
c
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
9
.
i
4
.
p
p
7
5
7
-
7
6
5
.
[
3
1
]
M
.
S
c
h
u
st
e
r
a
n
d
K
.
K
.
P
a
l
i
w
a
l
,
“
B
i
d
i
r
e
c
t
i
o
n
a
l
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
S
i
g
n
a
l
Pro
c
e
ss
i
n
g
,
v
o
l
.
4
5
,
n
o
.
1
1
,
p
p
.
2
6
7
3
–
2
6
8
1
,
1
9
9
7
,
d
o
i
:
1
0
.
1
1
0
9
/
7
8
.
6
5
0
0
9
3
.
[
3
2
]
K
.
C
h
o
e
t
a
l
.
,
“
Le
a
r
n
i
n
g
p
h
r
a
se
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
u
s
i
n
g
R
N
N
e
n
c
o
d
e
r
–
d
e
c
o
d
e
r
f
o
r
st
a
t
i
st
i
c
a
l
m
a
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
1
4
C
o
n
f
e
re
n
c
e
o
n
Em
p
i
r
i
c
a
l
M
e
t
h
o
d
s
i
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pr
o
c
e
ssi
n
g
(
EM
N
L
P)
,
S
t
r
o
u
d
sb
u
r
g
,
P
A
,
U
S
A
:
A
ss
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
s
t
i
c
s,
2
0
1
4
,
p
p
.
1
7
2
4
–
1
7
3
4
,
d
o
i
:
1
0
.
3
1
1
5
/
v
1
/
D
1
4
-
1
1
7
9
.
[
3
3
]
Y
.
B
e
n
g
i
o
,
R
.
D
u
c
h
a
r
m
e
,
a
n
d
P
.
V
i
n
c
e
n
t
,
“
A
n
e
u
r
a
l
p
r
o
b
a
b
i
l
i
st
i
c
l
a
n
g
u
a
g
e
mo
d
e
l
,
”
A
d
v
a
n
c
e
s
i
n
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
s
t
e
m
s
,
2
0
0
1
.
[
3
4
]
J.
P
e
n
n
i
n
g
t
o
n
,
R
.
S
o
c
h
e
r
,
a
n
d
C
.
M
a
n
n
i
n
g
,
“
G
l
o
v
e
:
g
l
o
b
a
l
v
e
c
t
o
r
s
f
o
r
w
o
r
d
r
e
p
r
e
s
e
n
t
a
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
1
4
C
o
n
f
e
r
e
n
c
e
o
n
Em
p
i
r
i
c
a
l
M
e
t
h
o
d
s
i
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pr
o
c
e
ssi
n
g
(
E
MN
L
P)
,
S
t
r
o
u
d
s
b
u
r
g
,
P
A
,
U
S
A
:
A
ss
o
c
i
a
t
i
o
n
f
o
r
C
o
mp
u
t
a
t
i
o
n
a
l
Li
n
g
u
i
s
t
i
c
s,
2
0
1
4
,
p
p
.
1
5
3
2
–
1
5
4
3
,
d
o
i
:
1
0
.
3
1
1
5
/
v
1
/
D
1
4
-
1
1
6
2
.
[
3
5
]
S
.
N
i
e
ß
e
n
a
n
d
H
.
N
e
y
,
“
I
mp
r
o
v
i
n
g
S
M
T
q
u
a
l
i
t
y
w
i
t
h
mo
r
p
h
o
-
s
y
n
t
a
c
t
i
c
a
n
a
l
y
si
s
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
1
8
t
h
c
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
l
i
n
g
u
i
st
i
c
s
,
M
o
r
r
i
s
t
o
w
n
,
N
J
,
U
S
A
:
A
ss
o
c
i
a
t
i
o
n
f
o
r
C
o
mp
u
t
a
t
i
o
n
a
l
Li
n
g
u
i
st
i
c
s
,
2
0
0
0
,
d
o
i
:
1
0
.
3
1
1
5
/
9
9
2
7
3
0
.
9
9
2
8
0
9
.
[
3
6
]
P
.
K
o
e
h
n
a
n
d
K
.
K
n
i
g
h
t
,
“
Em
p
i
r
i
c
a
l
met
h
o
d
s
f
o
r
c
o
mp
o
u
n
d
s
p
l
i
t
t
i
n
g
,
”
i
n
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
t
e
n
t
h
c
o
n
f
e
r
e
n
c
e
o
n
Eu
r
o
p
e
a
n
c
h
a
p
t
e
r
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
-
EA
C
L
’
0
3
,
M
o
r
r
i
s
t
o
w
n
,
N
J,
U
S
A
:
A
sso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
s
t
i
c
s,
2
0
0
3
,
d
o
i
:
1
0
.
3
1
1
5
/
1
0
6
7
8
0
7
.
1
0
6
7
8
3
3
.
[
3
7
]
S
.
V
i
r
p
i
o
j
a
,
J.
J
.
V
ä
y
r
y
n
e
n
,
M
.
C
r
e
u
t
z
,
a
n
d
M
.
S
a
d
e
n
i
e
m
i
,
“
M
o
r
p
h
o
l
o
g
y
-
a
w
a
r
e
st
a
t
i
st
i
c
a
l
m
a
c
h
i
n
e
t
r
a
n
sl
a
t
i
o
n
b
a
se
d
o
n
m
o
r
p
h
s
i
n
d
u
c
e
d
i
n
a
n
u
n
su
p
e
r
v
i
s
e
d
ma
n
n
e
r
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
M
a
c
h
i
n
e
T
r
a
n
s
l
a
t
i
o
n
S
u
m
m
i
t
XI
,
p
p
.
4
9
1
–
4
9
8
,
2
0
0
7
.
[
3
8
]
D
.
S
t
a
l
l
a
r
d
,
J
.
D
e
v
l
i
n
,
M
.
K
a
y
ser,
Y
.
K
.
Le
e
,
a
n
d
R
.
B
a
r
z
i
l
a
y
,
“
U
n
s
u
p
e
r
v
i
s
e
d
m
o
r
p
h
o
l
o
g
y
r
i
v
a
l
s
s
u
p
e
r
v
i
se
d
m
o
r
p
h
o
l
o
g
y
f
o
r
A
r
a
b
i
c
M
T,
”
i
n
5
0
t
h
A
n
n
u
a
l
Me
e
t
i
n
g
o
f
t
h
e
A
sso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
s
t
i
c
s,
A
C
L
2
0
1
2
-
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
C
o
n
f
e
re
n
c
e
,
p
p
.
3
2
2
–
3
2
7
,
2
0
1
2
.
[
3
9
]
P
.
Y
.
Ts
a
n
n
,
Y
.
K
.
H
o
o
i
,
M
.
F
.
H
a
ss
a
n
,
a
n
d
M
.
T
.
Y
.
W
o
o
i
,
“
L
e
a
d
i
n
g
s
e
n
t
e
n
c
e
n
e
w
s
T
e
x
t
R
a
n
k
,
”
i
n
2
0
2
1
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
t
e
l
l
i
g
e
n
t
C
y
b
e
rn
e
t
i
c
s
T
e
c
h
n
o
l
o
g
y
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
(
I
C
I
C
y
T
A)
,
I
EEE,
D
e
c
.
2
0
2
1
,
p
p
.
9
2
–
95
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
C
y
TA
5
3
7
1
2
.
2
0
2
1
.
9
6
8
9
1
8
6
.
[
4
0
]
L.
W
.
L
e
e
,
H
.
M
.
L
o
w
,
a
n
d
A
.
R
.
M
o
h
a
m
e
d
,
“
A
c
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
o
f
w
o
r
d
s
t
r
u
c
t
u
r
e
s
i
n
M
a
l
a
y
a
n
d
E
n
g
l
i
s
h
s
t
o
r
y
b
o
o
k
s
s
o
c
i
a
l
sci
e
n
c
e
s
a
n
d
h
u
ma
n
i
t
i
e
s
a
c
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
o
f
w
o
r
d
st
r
u
c
t
u
r
e
s
i
n
M
a
l
a
y
a
n
d
E
n
g
l
i
sh
c
h
i
l
d
r
e
n
’
s
s
t
o
r
i
e
s,”
Pe
r
t
a
n
i
k
a
J
o
u
rn
a
l
o
f
S
o
c
i
a
l
S
c
i
e
n
c
e
,
p
p
.
6
7
–
8
4
,
2
0
1
3
.
[
4
1
]
L.
W
.
L
e
e
,
“
D
e
v
e
l
o
p
me
n
t
a
n
d
v
a
l
i
d
a
t
i
o
n
o
f
a
r
e
a
d
i
n
g
-
r
e
l
a
t
e
d
a
ss
e
ssm
e
n
t
b
a
t
t
e
r
y
i
n
M
a
l
a
y
f
o
r
t
h
e
p
u
r
p
o
s
e
o
f
d
y
s
l
e
x
i
a
a
ssessm
e
n
t
,
”
An
n
a
l
s
o
f
D
y
sl
e
x
i
a
,
v
o
l
.
5
8
,
n
o
.
1
,
p
p
.
3
7
–
5
7
,
J
u
n
.
2
0
0
8
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
1
8
8
1
-
0
0
7
-
0
0
1
1
-
0.
[
4
2
]
S
.
S
.
-
S
h
w
a
r
t
z
a
n
d
T.
Z
h
a
n
g
,
“
A
c
c
e
l
e
r
a
t
e
d
p
r
o
x
i
ma
l
s
t
o
c
h
a
s
t
i
c
d
u
a
l
c
o
o
r
d
i
n
a
t
e
a
s
c
e
n
t
f
o
r
r
e
g
u
l
a
r
i
z
e
d
l
o
s
s
mi
n
i
mi
z
a
t
i
o
n
,
”
Ma
t
h
e
m
a
t
i
c
a
l
Pr
o
g
ra
m
m
i
n
g
,
v
o
l
.
1
5
5
,
p
p
.
1
0
5
–
1
4
5
,
Ja
n
.
2
0
1
6
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
1
0
7
-
0
1
4
-
0
8
3
9
-
0.
[
4
3
]
V
.
A
sh
i
sh
,
“
A
t
t
e
n
t
i
o
n
i
s
a
l
l
y
o
u
n
e
e
d
,
”
A
d
v
a
n
c
e
s
i
n
n
e
u
r
a
l
i
n
f
o
rm
a
t
i
o
n
p
ro
c
e
ssi
n
g
s
y
st
e
m
s
,
v
o
l
.
3
0
,
2
0
1
7
.
[
4
4
]
H
.
C
h
o
i
,
K
.
C
h
o
,
a
n
d
Y
.
B
e
n
g
i
o
,
“
F
i
n
e
-
g
r
a
i
n
e
d
a
t
t
e
n
t
i
o
n
mec
h
a
n
i
sm
f
o
r
n
e
u
r
a
l
ma
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
,
”
N
e
u
ro
c
o
m
p
u
t
i
n
g
,
v
o
l
.
2
8
4
,
p
p
.
1
7
1
–
1
7
6
,
A
p
r
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m
.
2
0
1
8
.
0
1
.
0
0
7
.
[
4
5
]
Y
.
Y
u
,
K
.
A
d
u
,
N
.
Ta
s
h
i
,
P
.
A
n
o
k
y
e
,
X
.
W
a
n
g
,
a
n
d
M
.
A
.
A
y
i
d
z
o
e
,
“
R
M
A
F
:
r
e
l
u
-
memr
i
st
o
r
-
l
i
k
e
a
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
f
o
r
d
e
e
p
l
e
a
r
n
i
n
g
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
8
,
p
p
.
7
2
7
2
7
–
7
2
7
4
1
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
0
.
2
9
8
7
8
2
9
.
[
4
6
]
T.
S
z
a
n
d
a
ł
a
,
“
R
e
v
i
e
w
a
n
d
c
o
m
p
a
r
i
so
n
o
f
c
o
mm
o
n
l
y
u
s
e
d
a
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
s
f
o
r
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
i
n
S
t
u
d
i
e
s
i
n
C
o
m
p
u
t
a
t
i
o
n
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
2
0
2
1
,
p
p
.
2
0
3
–
2
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
9
8
1
-
15
-
5
4
9
5
-
7
_
1
1
.
[
4
7
]
Y
.
T
.
P
h
u
a
,
S
.
N
a
v
a
r
a
t
n
a
m
,
C
.
-
M
.
K
a
n
g
,
a
n
d
W
.
-
S
.
C
h
e
,
“
S
e
q
u
e
n
c
e
-
to
-
s
e
q
u
e
n
c
e
n
e
u
r
a
l
m
a
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
f
o
r
E
n
g
l
i
s
h
-
M
a
l
a
y
,
”
I
A
E
S
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
I
J
-
A
I
)
,
v
o
l
.
1
1
,
n
o
.
2
,
p
p
.
6
5
8
–
6
6
5
,
J
u
n
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
1
.
i
2
.
p
p
6
5
8
-
6
6
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.