I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
24
,
No
.
3
,
Dec
em
b
er
2
0
2
1
,
p
p
.
1
8
1
4
~
1
8
2
2
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
24
.i
3
.
pp
1
8
1
4
-
1
8
2
2
1814
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Ama
zig
h
pa
rt
-
of
-
speech
tag
g
ing
wit
h ma
chine
learni
ng
and
deep learning
O
t
m
a
n M
a
a
ro
uf
1
,
Ra
chid E
l
Ay
a
chi
2
,
M
o
ha
m
ed
B
ini
z
3
1,
2
De
p
a
rtme
n
t
o
f
C
o
m
p
u
ter S
c
ien
c
e
,
F
a
c
u
lt
y
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
o
lo
g
y
,
S
u
lt
a
n
M
o
u
lay
S
li
m
a
n
e
Un
iv
e
rsit
y
,
Be
n
i
M
e
ll
a
l,
M
o
r
o
c
c
o
3
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
P
o
ly
d
isc
ip
l
in
a
ry
,
F
a
c
u
lt
y
S
u
lt
a
n
M
o
u
lay
,
S
li
m
a
n
e
Un
i
v
e
rsity
,
Be
n
i
M
e
ll
a
l,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
1
1
,
2
0
2
1
R
ev
is
ed
Oct
14
,
2
0
2
1
Acc
ep
ted
Oct
27
,
2
0
2
1
Na
tu
ra
l
lan
g
u
a
g
e
p
ro
c
e
ss
in
g
(
N
LP
)
is
a
p
a
rt
o
f
a
rti
ficia
l
i
n
telli
g
e
n
c
e
th
a
t
d
isse
c
ts,
c
o
m
p
re
h
e
n
d
s,
a
n
d
c
h
a
n
g
e
s
c
o
m
m
o
n
d
iale
c
ts
with
c
o
m
p
u
ters
i
n
c
o
m
p
o
se
d
a
n
d
s
p
o
k
e
n
se
tt
i
n
g
s.
A
t
th
a
t
p
o
in
t
i
n
sc
rip
ts.
G
ra
m
m
a
ti
c
a
l
fe
a
tu
re
s
p
a
rt
-
of
-
sp
e
e
c
h
(P
OS)
a
ll
o
w m
a
rk
i
n
g
th
e
wo
r
d
a
s p
e
r
it
s sta
tem
e
n
t.
W
e
fin
d
in
th
e
li
tera
tu
re
th
a
t
P
OS
is
u
se
d
i
n
a
fe
w
d
iale
c
ts,
in
p
a
rti
c
u
lar:
F
re
n
c
h
a
n
d
En
g
li
s
h
.
T
h
is
p
a
p
e
r
i
n
v
e
sti
g
a
tes
th
e
a
tt
e
n
ti
o
n
-
b
a
se
d
l
o
n
g
sh
o
r
t
-
term
m
e
m
o
ry
(LS
TM
)
n
e
two
rk
s
a
n
d
sim
p
le
re
c
u
rre
n
t
n
e
u
ra
l
n
e
two
rk
(RNN
)
in
Ti
fi
n
a
g
h
P
OS
tag
g
in
g
wh
e
n
it
is
c
o
m
p
a
re
d
to
c
o
n
d
it
io
n
a
l
ra
n
d
o
m
fiel
d
s
(CRF
)
a
n
d
d
e
c
isio
n
tree
.
Th
e
a
tt
ra
c
ti
v
e
n
e
s
s
o
f
LS
TM
n
e
tw
o
rk
s
is
th
e
ir
stre
n
g
th
i
n
m
o
d
e
li
n
g
lo
n
g
-
d
istan
c
e
d
e
p
e
n
d
e
n
c
ies
.
Th
e
e
x
p
e
rime
n
t
re
su
lt
s
sh
o
w
th
a
t
LS
TM
n
e
two
r
k
s
p
e
rfo
rm
b
e
tt
e
r
t
h
a
n
RNN
,
CRF
a
n
d
d
e
c
isio
n
tree
th
a
t
h
a
s
a
n
e
a
r
p
e
rfo
rm
a
n
c
e
.
K
ey
w
o
r
d
s
:
Am
az
ig
h
lan
g
u
ag
e
C
o
n
d
itio
n
al
r
an
d
o
m
f
ield
s
Dec
is
io
n
tr
ee
Dee
p
lear
n
in
g
Ma
ch
in
e
lear
n
in
g
Par
t o
f
s
p
ee
ch
T
if
in
ag
h
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
:
Otm
an
Ma
ar
o
u
f
Dep
ar
tm
en
t o
f
co
m
p
u
ter
s
cien
ce
Su
ltan
Mo
u
lay
Sli
m
an
e
Un
iv
e
r
s
ity
,
B
en
i M
ellal,
Mo
r
o
cc
o
E
m
ail: m
aa
r
o
u
f
.
o
tm
an
9
4
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Par
t
-
of
-
s
p
ee
ch
(
POS)lab
elin
g
is
th
e
way
to
war
d
d
o
lin
g
o
u
t
g
r
am
m
atica
l
f
ea
tu
r
e
m
ar
k
er
s
to
ea
ch
wo
r
d
in
an
in
f
o
r
m
atio
n
al
te
x
t
in
d
if
f
er
en
t
lan
g
u
ag
es
an
d
d
ialec
ts
.
[
1
]
T
h
e
co
n
tr
ib
u
tio
n
to
a
lab
elin
g
ca
lcu
latio
n
is
a
s
u
cc
ess
io
n
o
f
(
to
k
en
ized
)
wo
r
d
s
an
d
a
la
b
el
s
et,
an
d
th
e
y
ield
is
an
ar
r
an
g
e
m
en
t
o
f
lab
els,
o
n
e
f
o
r
ea
ch
to
k
en
[
2
]
.
T
h
u
s
ly
,
POS
tag
g
er
s
ar
e
an
im
p
o
r
t
b
u
g
m
o
d
u
le
f
o
r
g
ig
an
tic
o
p
e
n
ap
p
licatio
n
s
,
f
o
r
in
s
tan
ce
,
th
e
q
u
esti
o
n
s
-
tak
in
g
n
o
tes
o
f
s
y
s
tem
s
,
in
f
o
r
m
atio
n
ex
tr
ac
tio
n
,
i
n
f
o
r
m
atio
n
r
ec
u
p
er
atio
n
,
m
ac
h
i
n
e
u
n
d
er
s
t
an
d
i
n
g
.
T
h
e
y
ca
n
b
e
u
s
ed
in
v
ar
io
u
s
ap
p
licatio
n
s
,
f
o
r
in
s
tan
ce
,
s
u
b
s
tan
ce
to
talk
o
r
lik
e
a
p
r
ep
r
o
ce
s
s
o
r
f
o
r
a
p
ar
s
er
;
th
e
p
a
r
s
er
ca
n
i
m
p
r
o
v
e
,
in
an
y
ca
s
e,
d
y
n
am
ic
ally
ex
p
en
s
iv
e.
I
n
t
h
is
p
ap
er
,
we
d
ec
id
ed
to
f
o
c
u
s
o
n
th
e
POS
n
am
in
g
f
o
r
th
e
Am
az
ig
h
lan
g
u
a
g
e
.
T
h
ese
d
ay
s
,
POS
T
ag
g
in
g
is
f
in
is
h
ed
with
r
esp
ec
t
to
co
m
p
u
tatio
n
al
s
em
an
tics
u
s
in
g
a
c
o
u
p
le
o
f
co
u
n
ts
b
y
a
to
n
o
f
clea
r
m
ar
k
s
.
POS
-
lab
eli
n
g
f
i
g
u
r
in
g
f
all
i
n
to
th
r
ee
s
p
ec
if
ic
s
o
cial
af
f
air
s
:
r
u
le
-
b
ased
,
q
u
an
tifia
b
le,
a
n
d
h
y
b
r
id
-
b
ased
tag
g
er
s
.
A
s
tan
d
ar
d
-
ba
s
ed
tag
g
er
u
s
es
ety
m
o
lo
g
ical
r
u
les
to
co
n
s
ig
n
th
e
co
r
r
ec
t
m
ar
k
s
to
th
e
wo
r
d
s
in
th
e
s
en
ten
ce
o
r
r
ec
o
r
d
.
Au
th
en
tic
Par
t
o
f
Sp
ee
ch
tag
g
er
r
elies
u
p
o
n
t
h
e
p
r
o
b
ab
ilit
ies
o
f
o
cc
asio
n
s
o
f
wo
r
d
s
f
o
r
a
g
iv
en
e
x
p
licit
tag
th
r
o
u
g
h
a
d
ec
is
io
n
tr
ee
an
d
co
n
d
itio
n
al
r
an
d
o
m
f
ield
s
(
C
R
F)
ap
p
r
o
ac
h
.
C
r
ea
m
-
b
ased
Par
t
o
f
Sp
ee
ch
tag
g
er
is
a
b
len
d
o
f
a
r
u
le
-
b
ased
s
y
s
tem
an
d
f
ac
tu
al
p
h
ilo
s
o
p
h
y
.
Sy
n
tactic
s
tr
u
ctu
r
e
n
am
in
g
is
a
h
u
g
e
g
ad
g
et
o
f
n
o
r
m
al
lan
g
u
ag
e
ar
r
an
g
em
e
n
t.
I
t
is
u
s
ed
i
n
a
c
o
u
p
le
o
f
Natu
r
al
L
a
n
g
u
a
g
e'
s
r
ea
d
in
ess
-
b
ased
p
r
o
g
r
am
m
in
g
ex
ec
u
tio
n
s
.
Fo
r
all
n
at
u
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P)
task
s
lik
e
lan
g
u
ag
e
s
tr
u
ctu
r
e,
c
h
ec
k
er
m
ac
h
i
n
e,
a
n
d
tr
a
n
s
latio
n
,
th
e
p
r
ec
is
io
n
d
ep
e
n
d
s
o
n
th
e
ac
cu
r
ac
y
o
f
th
e
p
ar
t
o
f
s
p
ee
ch
tag
g
er
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
ma
z
ig
h
p
a
r
t
-
of
-
s
p
ee
ch
ta
g
g
i
n
g
w
ith
ma
ch
in
e
lea
r
n
i
n
g
a
n
d
d
ee
p
lea
r
n
in
g
(
Otma
n
Ma
a
r
o
u
f)
1815
T
ag
g
er
ac
ce
p
ts
a
c
r
itical
o
cc
u
p
atio
n
in
talk
af
f
i
r
m
atio
n
,
n
o
r
m
al
lan
g
u
ag
e
p
a
r
s
in
g
an
d
i
n
f
o
r
m
atio
n
r
ec
u
p
er
atio
n
.
T
h
is
u
n
d
er
tak
i
n
g
s
ee
k
s
af
ter
th
e
m
ea
s
u
r
ab
le
b
ased
tech
n
iq
u
e
f
o
r
POS
T
ag
g
in
g
,
m
u
ch
m
o
r
e
ex
p
r
ess
ly
th
e
C
R
F
Mo
d
el
o
f
th
e
r
ea
l
s
y
s
tem
[
3
]
.
As
o
f
late,
n
eu
r
al
o
r
g
a
n
izatio
n
s
h
a
v
e
b
ee
n
ac
q
u
ir
i
n
g
n
o
to
r
iety
in
th
e
f
ield
o
f
co
m
p
u
ter
ized
r
ea
s
o
n
in
g
.
T
h
e
h
ea
d
way
is
b
ec
au
s
e
o
f
th
e
d
is
co
v
e
r
y
in
th
e
ca
lcu
latio
n
s
th
at
lear
n
an
d
p
er
ce
iv
e
ex
tr
e
m
ely
co
m
p
lex
e
x
am
p
les
u
tili
zin
g
p
r
o
f
o
u
n
d
lay
er
s
o
f
n
eu
r
al
o
r
g
an
izatio
n
s
o
r
n
o
r
m
ally
k
n
o
wn
as
t
h
e
p
r
o
f
o
u
n
d
n
e
u
r
al
o
r
g
an
izatio
n
s
d
e
ep
n
e
u
r
al
n
etwo
r
k
(
DNN)
[
4
]
,
f
u
r
th
er
m
o
r
e,
th
e
p
r
esen
tatio
n
o
f
v
a
r
io
u
s
k
in
d
s
o
f
n
e
u
r
al
o
r
g
an
izatio
n
,
n
am
el
y
,
co
n
v
o
lu
tio
n
al
n
e
u
r
al
o
r
g
an
i
za
tio
n
an
d
r
ep
etitiv
e
n
eu
r
al
o
r
g
an
izatio
n
(
R
NN)
.
Fo
r
ex
am
p
le,
co
n
v
o
lu
ti
o
n
al
n
eu
r
al
o
r
g
a
n
izatio
n
s
,
wh
ich
ar
e
a
u
n
iq
u
e
k
i
n
d
o
f
f
ee
d
-
f
o
r
war
d
n
eu
r
al
o
r
g
an
iz
atio
n
s
with
two
m
ea
s
u
r
em
e
n
t
o
r
g
an
izatio
n
s
,
h
av
e
d
em
o
n
s
tr
ated
co
l
o
s
s
al
ex
ac
tn
ess
in
g
r
o
u
p
s
p
ictu
r
es
th
r
o
u
g
h
n
ea
r
b
y
o
p
en
f
ield
s
,
s
h
ar
ed
lo
ad
s
,
p
o
o
lin
g
,
f
r
o
m
s
tr
aig
h
tf
o
r
war
d
tr
an
s
cr
ib
ed
d
ig
it
ac
k
n
o
wled
g
m
en
t
to
m
o
r
e
p
er
p
le
x
in
g
f
a
ce
ac
k
n
o
wled
g
m
e
n
t.
I
n
th
e
d
em
o
n
s
tr
atin
g
o
f
co
n
s
ec
u
tiv
e
e
x
am
p
les,
lik
e
,
p
h
o
n
em
e
ac
k
n
o
wled
g
m
e
n
t
p
r
o
g
r
am
m
ed
d
is
co
u
r
s
e
ac
k
n
o
wled
g
m
en
t,
d
is
co
u
r
s
e
co
m
b
in
atio
n
,
d
is
co
u
r
s
e
tr
an
s
la
tio
n
,
ch
atb
o
t,
an
d
n
u
m
er
o
u
s
o
t
h
er
s
,
R
NN
o
r
th
e
m
o
r
e
s
p
ec
if
i
c
s
o
r
t
o
f
R
NN
an
d
th
e
lo
n
g
-
s
h
o
r
t
ter
m
m
em
o
r
y
(
L
STM
)
n
etwo
r
k
s
[5
]
,
[
6]
h
av
e
d
em
o
n
s
tr
ated
to
b
e
s
u
p
e
r
io
r
t
o
lar
g
e
n
u
m
b
er
s
o
f
th
e
co
n
v
en
tio
n
al
m
eth
o
d
o
l
o
g
i
es
.
T
h
is
p
a
p
er
p
r
esen
ts
a
s
im
ilar
in
v
esti
g
atio
n
o
f
th
r
ee
s
tr
ate
g
ies
to
tak
e
ca
r
e
o
f
th
e
is
s
u
e
o
f
Am
az
i
g
h
g
r
am
m
a
tical
f
ea
tu
r
e
(
POS)
lab
elin
g
.
T
h
ese
tech
n
iq
u
es
a
r
e
L
STM
o
r
g
an
izatio
n
s
,
C
R
F,
an
d
d
ec
is
io
n
tr
ee
s
.
T
h
e
g
o
al
i
s
to
an
aly
ze
th
e
ex
h
i
b
itio
n
o
f
th
e
p
r
esen
t
s
tatu
s
o
f
L
STM
o
r
g
an
izatio
n
s
w
h
ile
co
n
tr
asted
with
C
R
F
an
d
Dec
is
io
n
tr
ee
s
in
POS
lab
elin
g
.
POS
lab
elin
g
is
a
lan
g
u
ag
e
-
p
r
ep
ar
in
g
task
th
at
allo
ca
ted
a
POS
tag
to
ea
ch
wo
r
d
in
a
s
en
ten
ce
[
4
]
.
C
o
n
c
er
n
in
g
th
e
o
r
g
a
n
izatio
n
o
f
p
a
p
er
,
th
er
e
a
r
e
th
r
ee
s
ec
tio
n
s
,
in
th
e
f
ir
s
t
s
ec
tio
n
w
e
talk
ab
o
u
t
th
e
d
if
f
er
en
t
ap
p
r
o
ac
h
es
u
s
ed
in
th
e
Am
az
ig
h
p
ar
t
o
f
s
p
ee
ch
,
in
t
h
e
s
ec
o
n
d
s
ec
tio
n
we
d
is
p
u
te
th
e
m
eth
o
d
o
lo
g
y
o
f
th
e
ar
ticle
ta
lk
in
g
a
b
o
u
t
th
e
Am
az
ig
h
lan
g
u
ag
e,
t
h
e
Am
az
ig
h
t
ag
s
et
an
d
th
e
co
r
p
u
s
u
s
in
g
i
n
th
is
s
tu
d
y
.
T
h
e
t
h
ir
d
s
ec
tio
n
is
d
iv
id
ed
in
th
r
ee
p
ar
ts
,
th
e
f
ir
s
t
p
ar
t
d
escr
ib
es
th
e
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
(
C
R
F
an
d
d
ec
is
io
n
tr
ee
)
,
th
e
s
ec
o
n
d
p
ar
t
p
r
esen
ts
th
e
ar
ch
itectu
r
e
o
f
th
e
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
a
n
d
lo
n
g
-
s
h
o
r
t
ter
m
m
em
o
r
y
,
b
u
t
i
n
th
e
last
p
ar
t,
th
e
o
b
tain
ed
r
e
s
u
lts
ar
e
d
is
cu
s
s
ed
an
d
th
e
c
o
n
clu
s
io
n
is
g
iv
e
n
.
2.
RE
L
AT
E
D
WO
RK
S
I
n
t
h
e
d
o
m
ai
n
o
f
POS
l
ab
eli
n
g
,
n
u
m
e
r
o
u
s
i
n
v
es
ti
g
at
io
n
s
h
a
v
e
b
e
en
m
a
d
e
.
I
t
ar
r
i
v
ed
a
t
i
n
c
r
ed
ib
le
d
e
g
r
ee
s
o
f
ex
ec
u
ti
o
n
u
s
i
n
g
d
is
cr
i
m
i
n
a
ti
v
e
m
o
d
els
,
f
o
r
e
x
am
p
le,
g
r
e
ates
t
e
n
t
r
o
p
y
m
o
d
els
(
M
ax
E
n
t)
[
7
]
,
b
o
ls
t
er
v
e
ct
o
r
m
a
c
h
i
n
es
(
B
VM
)
[
8
]
,
s
u
p
p
o
r
t
v
e
ct
o
r
m
a
c
h
i
n
e
(
SVM
)
is
a
n
e
x
a
m
p
le
o
f
s
u
p
e
r
v
is
e
d
l
ea
r
n
in
g
al
g
o
r
i
th
m
s
th
a
t
h
is
f
u
n
cti
o
n
is
t
o
f
i
n
d
a
h
y
p
er
p
l
a
n
e
t
h
at
s
e
p
ar
ates
d
at
a
i
n
tw
o
cl
ass
es
[
9
]
,
o
r
M
ar
k
o
v
C
R
F
[
1
0
]
,
[
1
1
]
.
Am
o
n
g
s
t
o
c
h
ast
ic
m
o
d
els
,
b
i
g
r
a
m
an
d
tr
i
g
r
am
h
id
d
en
M
ar
k
o
v
m
o
d
e
ls
(
HM
M)
a
r
e
v
e
r
y
p
r
o
m
in
e
n
t.
D
y
n
a
m
it
e
is
a
g
e
n
e
r
al
ly
u
til
ize
d
s
t
o
c
h
as
tic
tr
ig
r
a
m
HM
M
ta
g
g
e
r
w
h
i
ch
u
tili
ze
s
a
s
u
f
fi
x
ex
am
in
ati
o
n
s
y
s
te
m
to
ass
ess
lex
ic
al
p
r
o
b
ab
ilit
ies
f
o
r
o
b
s
c
u
r
e
t
o
k
e
n
s
d
ep
en
d
en
t
o
n
p
r
o
p
e
r
t
ies
o
f
t
h
e
w
o
r
d
s
i
n
th
e
p
r
e
p
a
r
ati
o
n
c
o
r
p
u
s
w
h
ic
h
s
h
a
r
e
t
h
e
e
q
u
iv
ale
n
t
s
u
f
fi
x
[
1
2
]
.
T
h
e
im
p
r
o
v
e
m
e
n
t
o
f
a
s
to
c
h
asti
c
ta
g
g
e
r
r
e
q
u
i
r
es
a
lo
t
o
f
c
o
m
m
e
n
t
o
n
c
o
n
te
n
t
.
Sto
ch
asti
c
ta
g
g
e
r
s
w
it
h
o
v
er
9
5
%
w
o
r
d
-
le
v
el
e
x
a
ct
n
ess
h
a
v
e
b
e
e
n
p
r
o
d
u
ce
d
i
n
E
n
g
l
is
h
,
Ger
m
a
n
an
d
o
t
h
e
r
E
u
r
o
p
ea
n
d
i
ale
cts,
i
n
w
h
ic
h
h
u
g
e
n
a
m
e
d
i
n
f
o
r
m
ati
o
n
is
a
cc
ess
ib
le
.
At
t
h
at
p
o
in
t,
c
h
o
i
ce
t
r
e
es
h
a
v
e
b
e
e
n
u
ti
liz
ed
f
o
r
P
OS
la
b
eli
n
g
a
n
d
p
a
r
s
i
n
g
.
A
ch
o
ice
tr
ee
i
n
i
tia
te
d
f
r
o
m
la
b
el
e
d
c
o
r
p
o
r
a
w
as
u
s
ed
f
o
r
g
r
a
m
m
ati
ca
l
f
e
at
u
r
e
d
is
a
m
b
i
g
u
ati
o
n
.
F
o
r
Am
a
zi
g
h
POS
la
b
e
li
n
g
[
1
3
]
.
Ma
n
u
f
a
ct
u
r
e
d
a
POS
-
t
ag
g
e
r
f
o
r
A
m
az
ig
h
,
as
a
n
u
n
d
er
-
r
es
o
u
r
ce
d
la
n
g
u
a
g
e
.
T
h
e
in
f
o
r
m
a
ti
o
n
u
s
e
d
t
o
a
c
h
ie
v
e
th
e
wo
r
k
was
p
h
y
s
ica
ll
y
g
at
h
e
r
e
d
a
n
d
c
o
m
m
e
n
t
ed
o
n
.
T
o
h
el
p
i
n
c
r
e
ase
th
e
p
r
es
en
t
ati
o
n
o
f
t
h
e
t
a
g
g
er
,
t
h
e
y
u
t
iliz
ed
AI
m
et
h
o
d
s
(
SV
M
a
n
d
C
R
F)
a
n
d
d
if
f
e
r
e
n
t
ass
ets
o
r
a
p
p
a
r
at
u
s
es,
f
o
r
e
x
a
m
p
le
,
le
x
i
c
o
n
s
a
n
d
wo
r
d
d
i
v
is
io
n
d
e
v
ic
es
t
o
p
r
o
ce
s
s
t
h
e
c
o
n
ten
t
a
n
d
c
o
n
c
en
tr
ate
h
i
g
h
li
g
h
ts
s
ets
c
o
m
p
r
is
i
n
g
o
f
lex
ic
al
s
e
tti
n
g
a
n
d
c
h
a
r
a
cte
r
n
-
g
r
am
s
.
T
h
e
c
o
r
p
u
s
c
o
n
ta
in
e
d
2
0
,
0
0
0
to
k
en
s
a
n
d
was
u
tili
ze
d
t
o
p
r
e
p
a
r
e
t
h
ei
r
POS
-
t
ag
g
er
m
o
d
el
.
I
n
t
h
is
m
an
n
e
r
,
t
h
e
r
e
is
a
s
q
u
ee
zi
n
g
n
ee
d
t
o
b
u
i
ld
u
p
a
p
r
o
g
r
a
m
m
ed
p
ar
t
-
of
-
s
p
ee
c
h
ta
g
g
e
r
f
o
r
A
m
a
zi
g
h
.
U
n
d
e
r
t
h
e
cu
r
r
e
n
t
wo
r
k
,
a
c
h
ar
ac
t
e
r
r
ec
o
g
n
iti
o
n
s
y
s
te
m
is
p
r
ese
n
t
ed
f
o
r
r
ec
o
g
n
i
zi
n
g
E
n
g
lis
h
ch
ar
ac
te
r
s
e
x
t
r
a
cte
d
f
r
o
m
i
m
a
g
es
/g
r
ap
h
i
cs
e
m
b
e
d
d
e
d
te
x
t
d
o
c
u
m
en
ts
s
u
c
h
as
b
u
s
in
ess
ca
r
d
im
ag
es.
I
n
t
h
e
s
am
e
wa
y
,
we
f
in
d
[
1
4
]
,
t
h
a
t
a
p
p
lie
d
t
h
e
t
r
e
e
ta
g
g
er
o
n
A
m
a
zi
g
h
te
x
ts
s
h
o
w
th
at
p
r
o
v
id
es
o
v
e
r
al
l
ta
g
g
i
n
g
ac
cu
r
ac
y
o
f
9
3
.
1
9
%,
s
p
ec
ifi
ca
l
ly
,
9
4
.
1
0
%
o
n
k
n
o
w
n
w
o
r
d
s
an
d
7
0
.
2
9
%
o
n
u
n
k
n
o
w
n
w
o
r
d
s
.
Ap
p
ly
in
g
th
e
n
ew
NL
P
tech
n
i
q
u
es,
we
f
in
d
[
1
5
]
ar
e
u
s
ed
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
ST
M)
o
n
th
e
Am
az
ig
h
p
ar
t
o
f
s
p
ee
ch
in
th
e
ca
s
e
o
f
tex
ts
wr
itten
in
L
atin
ch
ar
ac
ter
s
,
th
ey
r
ea
ch
a
s
co
r
e
o
f
9
2
.
7
%
in
a
tr
ain
in
g
o
f
3
0
ep
o
ch
s
.
W
e
f
o
u
n
d
also
[
1
6
]
u
s
ed
C
R
F,
d
ec
i
s
io
n
tr
ee
,
an
d
lo
n
g
s
h
o
r
t
ter
m
m
em
o
r
y
,
th
ey
r
ea
ch
a
s
co
r
e
o
f
8
7
.
6
% u
s
in
g
DT
,
9
3
.
4
% b
y
C
R
F,
an
d
9
7
.
8
7
% in
a
tr
ain
in
g
o
f
1
0
ep
o
c
h
s
.
3.
M
E
T
H
O
DO
L
O
G
Y
3
.
1
.
A
m
a
zig
h
l
a
ng
ua
g
e
Be
f
i
r
s
t
,
Am
az
ig
h
l
an
g
u
a
g
e
is
o
n
e
o
f
th
e
Ha
m
it
o
-
Se
m
it
ic/
"A
f
r
o
-
As
iat
ic"
d
i
ale
cts
w
it
h
a
r
ic
h
t
em
p
l
ati
c
m
o
r
p
h
o
lo
g
y
.
I
n
s
em
an
tic
t
er
m
s
,
t
h
e
l
an
g
u
a
g
e
is
d
es
c
r
i
b
e
d
b
y
t
h
e
m
u
l
ti
p
li
ca
t
io
n
o
f
t
o
n
g
u
es
b
e
ca
u
s
e
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
24
,
No
.
3
,
Dec
em
b
er
2
0
2
1
:
1
8
1
4
-
1
8
2
2
1816
v
e
r
i
f
ia
b
l
e,
g
e
o
l
o
g
ica
l,
a
n
d
s
o
ci
o
l
in
g
u
is
t
ic
ele
m
e
n
ts
.
I
n
Mo
r
o
cc
o
,
o
n
e
m
a
y
r
ec
o
g
n
iz
e
t
h
r
ee
s
i
g
n
i
f
i
ca
n
t
v
e
r
n
ac
u
l
a
r
s
:
T
a
r
i
f
it
i
n
th
e
N
o
r
th
,
T
a
m
a
zi
g
h
t
in
t
h
e
m
id
d
l
e,
a
n
d
T
as
h
l
h
i
y
t
in
t
h
e
s
o
u
th
er
n
r
e
g
i
o
n
s
o
f
th
e
co
u
n
t
r
y
; h
al
f
o
f
th
e
M
o
r
o
cc
an
p
o
p
u
l
ac
e
s
p
ea
k
s
A
m
a
zi
g
h
y
et
as i
n
d
ica
te
d
b
y
t
h
e
l
ast l
e
g
is
la
t
iv
e
d
em
o
-
li
n
g
u
is
ti
c
in
f
o
r
m
at
io
n
b
y
2
0
0
4
,
t
h
e
Am
a
zig
h
la
n
g
u
ag
e
i
s
s
p
o
k
e
n
o
n
l
y
b
y
s
o
m
e
2
8
% o
f
t
h
e
M
o
r
o
c
ca
n
p
o
p
u
l
ati
o
n
(
a
r
o
u
n
d
10
m
ill
io
n
ci
tiz
e
n
s
)
,
d
em
o
n
s
t
r
ati
n
g
a
s
i
g
n
i
f
ic
a
n
t
d
i
m
i
n
is
h
i
n
g
o
f
its
u
s
e
.
Am
az
ig
h
s
tan
d
ar
d
is
atio
n
ca
n
'
t
b
e
ac
co
m
p
lis
h
ed
with
o
u
t
em
b
r
ac
in
g
a
p
r
ac
tical
p
r
o
ce
d
u
r
e
th
at
co
n
s
id
er
its
p
h
o
n
etic
v
ar
iety
.
All
th
in
g
s
co
n
s
id
er
ed
,
an
d
as
a
r
esu
lt
o
f
r
ec
o
r
d
ed
an
d
s
o
cial
r
ea
s
o
n
s
,
T
if
in
ag
h
e
h
as
b
ec
o
m
e
th
e
o
f
f
icial
s
cr
ip
t
f
o
r
th
e
wr
itin
g
Am
az
i
g
h
lan
g
u
ag
e
.
I
R
C
AM
k
ep
t
ju
s
t
r
elev
an
t
p
h
o
n
e
m
es
f
o
r
T
am
az
ig
h
t,
s
o
th
e
q
u
a
n
tity
o
f
th
e
s
eq
u
en
tial
p
h
o
n
etic
s
u
b
s
t
an
ce
s
is
3
3
,
y
et
Un
ico
d
e
co
d
es
ju
s
t
3
1
letter
s
in
ad
d
itio
n
to
a
m
o
d
if
ier
letter
to
s
h
ap
e
th
e
two
p
h
o
n
etic
u
n
its
:
ⴳ
ⵯ
(
g
ʷ
)
a
n
d
ⴽⵯ
(
k
ʷ
)
.
T
h
e
en
tir
e
s
co
p
e
o
f
T
if
in
ag
h
letter
s
is
p
ar
titi
o
n
ed
in
to
f
o
u
r
s
u
b
s
ets:
th
e
letter
s
u
tili
ze
d
b
y
I
R
C
AM
,
an
all
-
in
clu
s
iv
e
s
e
t
u
s
ed
b
y
I
R
C
AM
,
o
th
er
n
ew
T
if
i
n
ag
h
letter
s
b
ei
n
g
u
s
ed
,
a
n
d
s
o
m
e
a
u
th
en
tica
ted
cu
r
r
e
n
t
T
o
u
a
r
e
g
letter
s
.
T
h
e
n
u
m
b
er
r
ea
c
h
es
5
5
ch
ar
ac
te
r
s
[
1
7
]
.
Am
az
ig
h
NL
P p
r
esen
ts
m
an
y
ch
allen
g
e
s
f
o
r
r
esear
ch
er
s
.
I
ts
m
ajo
r
f
ea
tu
r
es a
r
e:
−
I
t
h
as
its
o
w
n
s
c
r
i
p
t:
t
h
e
T
ifin
a
g
h
w
h
i
ch
is
w
r
it
te
n
f
r
o
m
le
f
t
t
o
r
i
g
h
t
.
−
I
t
d
o
es
n
o
t
c
o
n
tai
n
u
p
p
e
r
c
a
s
e
.
−
L
i
k
e
o
t
h
e
r
n
a
tu
r
al
l
a
n
g
u
a
g
es,
Am
a
zi
g
h
p
r
ese
n
ts
f
o
r
NL
P
a
m
b
ig
u
i
ties
i
n
g
r
am
m
a
r
class
es
,
n
a
m
e
d
e
n
t
iti
es,
an
d
m
ea
n
i
n
g
.
Fo
r
e
x
am
p
l
e,
g
r
am
m
at
ica
ll
y
;
t
h
e
wo
r
d
(
ill
i)
d
ep
en
d
i
n
g
o
n
th
e
c
o
n
te
x
t
ca
n
m
ea
n
a
n
o
u
n
i
n
th
is
s
e
n
t
en
ce
(
tf
u
l
k
i
i
lli:
m
y
d
a
u
g
h
te
r
is
b
e
au
t
i
f
u
l)
o
r
a
v
e
r
b
i
n
t
h
is
s
e
n
te
n
c
e
(
u
r
i
lli
wa
lo
u
:
t
h
er
e
is
n
o
th
in
g
)
.
−
As
m
o
s
t
la
n
g
u
ag
es
w
h
o
s
e
r
es
ea
r
c
h
in
NL
P
is
n
ew
,
A
m
az
ig
h
is
n
o
t
en
d
o
we
d
wi
th
th
e
li
n
g
u
is
ti
c
r
eso
u
r
c
es
an
d
NL
P
t
o
o
ls
.
−
Am
az
ig
h
,
lik
e
m
o
s
t
o
f
th
e
lan
g
u
ag
es
wh
ich
h
a
v
e
o
n
ly
r
ec
en
tly
s
tar
ted
b
ei
n
g
in
v
esti
g
ated
f
o
r
NL
P,
s
till
s
u
f
f
er
f
r
o
m
th
e
s
ca
r
city
o
f
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
to
o
ls
an
d
r
es
o
u
r
ce
s
[1
8
]
.
3
.
2
.
A
m
a
zig
h t
a
g
s
et
C
h
ar
ac
ter
izin
g
th
e
ad
eq
u
ate
l
ab
el
s
et
is
a
m
id
d
le
en
d
ea
v
o
r
in
b
u
ild
i
n
g
a
cu
s
to
m
ized
POS
tag
g
er
.
I
t
tar
g
ets
d
escr
ib
in
g
a
m
ea
s
u
r
ab
l
e
lab
el
s
et
to
a
r
ea
s
o
n
ab
le
lev
e
l
o
f
war
r
an
ty
,
f
o
r
in
s
tan
ce
,
n
o
t
h
ig
h
ly
-
elab
o
r
ated
n
o
r
u
n
r
ea
s
o
n
ab
ly
s
h
allo
w
f
o
r
th
e
p
o
te
n
tial
jo
in
s
s
tr
u
ctu
r
es
th
at
will
u
s
e
it.
T
h
e
u
s
ed
co
r
p
u
s
co
n
tain
s
a
o
n
ce
-
o
v
er
o
f
co
m
p
o
s
itio
n
s
is
o
lated
f
r
o
m
a
co
llectio
n
o
f
s
o
u
r
c
es,
f
o
r
in
s
tan
ce
a
c
o
u
p
le
o
f
b
o
o
k
s
,
s
im
ilar
ly
as
s
p
ec
if
ic
wo
r
k
s
f
r
o
m
I
R
C
AM
's
s
ite.
W
e
h
ad
th
e
ch
o
ice
to
s
h
o
w
u
p
at
an
all
d
war
f
o
f
wo
r
d
s
in
a
way
th
at
is
b
etter
th
an
6
0
k
to
k
en
s
.
T
h
is
c
o
r
p
u
s
is
r
em
ar
k
ed
o
n
m
o
r
p
h
o
lo
g
ically
u
s
in
g
th
e
m
ar
k
s
et
in
tr
o
d
u
ce
d
in
[
19
]
.
3
.
3
.
Co
rpus
A
co
r
p
u
s
is
a
s
et
o
f
la
n
g
u
a
g
e
i
n
f
o
r
m
ati
o
n
t
h
a
t
is
c
h
o
s
e
n
a
n
d
c
o
o
r
d
i
n
at
e
d
b
y
u
n
eq
u
i
v
o
c
al
s
e
m
a
n
ti
c
s
tan
d
ar
d
s
t
o
f
ill
i
n
as
a
n
e
x
a
m
p
le
o
f
o
cc
u
p
ati
o
n
s
d
ec
id
e
d
on
a
l
a
n
g
u
ag
e.
G
e
n
e
r
al
ly
s
p
e
ak
in
g
,
a
c
o
r
p
u
s
c
o
n
tai
n
s
a
h
u
g
e
n
u
m
b
er
o
f
w
o
r
d
s
a
n
d
c
an
b
e
le
m
m
ati
ze
d
a
n
d
cla
r
i
f
i
e
d
wit
h
d
ata
a
b
o
u
t
t
h
e
g
r
am
m
a
tica
l
f
o
r
m
s
.
A
m
o
n
g
th
e
c
o
r
p
u
s
,
t
h
er
e
is
t
h
e
B
r
it
i
s
h
Na
ti
o
n
al
C
o
r
p
u
s
(
1
0
0
m
i
l
lio
n
w
o
r
d
s
)
a
n
d
t
h
e
A
m
e
r
i
ca
n
Na
ti
o
n
al
C
o
r
p
u
s
(
2
0
m
ill
io
n
w
o
r
d
s
)
.
A
r
ea
s
o
n
ab
l
e
c
o
r
p
u
s
w
o
u
l
d
g
i
v
e
a
wi
d
e
c
h
o
i
ce
o
f
v
a
r
i
o
u
s
k
in
d
s
o
f
wr
iti
n
g
a
n
d
f
r
o
m
d
i
f
f
er
e
n
t
s
o
u
r
ce
s
,
f
o
r
ex
am
p
l
e,
p
ap
e
r
s
,
b
o
o
k
s
,
r
e
f
e
r
e
n
c
e
b
o
o
k
s
o
r
t
h
e
we
b
.
F
o
r
th
e
M
o
r
o
cc
a
n
A
m
az
ig
h
lan
g
u
a
g
e,
it
was
v
e
r
y
h
a
r
d
t
o
fin
d
i
n
s
t
an
t
ass
ets
.
W
e
ca
n
s
im
p
l
y
m
a
k
e
r
ef
er
e
n
ce
t
o
t
h
e
p
h
y
s
ical
c
o
m
m
e
n
t
e
d
o
n
th
e
c
o
r
p
u
s
o
f
O
u
ta
h
aja
la
et
a
l
.
[
2
0
]
.
T
h
is
c
o
r
p
u
s
co
n
t
ai
n
s
2
0
k
wo
r
d
s
,
w
h
ic
h
is
wh
y
we
h
a
v
e
d
e
ci
d
e
d
t
o
u
s
e
it
in
th
is
s
t
u
d
y
wi
th
s
o
m
e
m
o
d
if
ica
t
io
n
s
.
−
W
e
e
n
r
ic
h
ed
t
h
e
c
o
r
p
u
s
wi
t
h
o
t
h
e
r
wo
r
d
s
,
w
e
c
o
ll
ec
t
ed
m
o
r
e
t
h
a
n
4
0
k
wo
r
d
s
f
r
o
m
M
o
r
r
o
c
an
a
g
en
c
y
o
f
p
r
ess
(
M
AP
)
n
ews
s
o
w
e
h
av
e
in
g
lo
b
al
c
o
r
p
u
s
m
o
r
e
th
an
6
0
k
wo
r
d
s
−
W
e
h
a
v
e
a
d
d
e
d
t
h
e
t
y
p
e
s
y
m
b
o
l
i
n
th
e
t
y
p
es
o
f
ta
g
s
s
i
g
n
if
y
i
n
g
th
e
s
p
e
ci
al
ch
a
r
ac
te
r
s
t
h
at
th
ey
a
r
e
co
n
s
i
d
e
r
e
d
p
u
n
c
tu
ati
o
n
in
t
h
e
co
r
p
u
s
r
ea
liz
e
d
b
y
O
u
t
a
h
aj
ala
;
−
W
e
h
a
v
e
el
im
in
ate
d
t
h
e
d
at
e
,
t
y
p
e
b
ec
au
s
e
it
is
an
e
n
t
it
y
;
−
W
e
u
s
e
T
i
f
i
n
a
g
h
c
h
a
r
a
cte
r
s
i
n
s
tea
d
o
f
L
a
ti
n
w
r
i
ti
n
g
.
T
h
e
a
d
v
a
n
ta
g
e
o
f
u
s
in
g
t
h
e
T
if
in
a
g
h
ch
a
r
ac
te
r
is
t
o
o
p
t
i
m
iz
e
t
h
e
p
r
o
g
r
am
b
y
a
v
o
i
d
i
n
g
t
h
e
u
s
e
o
f
tr
a
n
s
co
d
i
n
g
.
I
n
t
h
is
s
t
e
p
,
we
h
a
v
e
e
n
c
o
u
n
te
r
e
d
s
o
m
e
e
r
r
o
r
s
i
n
t
h
e
w
r
iti
n
g
o
f
th
e
w
o
r
d
s
i
n
L
at
i
n
;
th
er
e
i
s
a
ch
ar
ac
t
er
i
n
t
h
e
c
o
r
p
u
s
o
f
O
u
t
ah
aja
la
wr
itt
en
i
n
a
n
er
r
o
n
e
o
u
s
wa
y
li
k
e
(
ⵟ
=
T
ⵃ
=
H
ⵥ
=
Z
ⵄ
=
E
ⴹ
=
D
ⵕ
=
R
ⵚ
=
S
ⵖ
=
G
Gw=
ⴳⵯ
)
(
T
a
b
l
e
1
)
,
b
u
t
th
e
co
r
r
e
ct
e
q
u
iv
al
en
ce
o
f
t
h
is
c
h
a
r
a
ct
er
is
(
ⵟ
=
ṭ
ⵃ
=
ḥ
ⵥ
=
ẓ
ⵄ
=
ɛ
ⴹ
=
ḍ
ⵕ
=
ṛ
ⵚ
=
ṣ
ⵖ
=
Ɣ
ⴳⵯ
=
g
w
)
.
Af
te
r
t
h
e
t
r
a
n
s
lati
o
n
i
n
t
o
T
i
f
i
n
a
g
h
c
h
a
r
a
ct
er
,
it
g
i
v
es
u
s
m
ea
n
i
n
g
l
ess
wo
r
d
s
o
r
w
o
r
d
s
w
it
h
a
d
i
f
f
er
e
n
t
m
e
an
in
g
f
r
o
m
t
h
e
o
r
ig
i
n
.
T
h
e
T
a
b
l
e
1
cit
es
s
o
m
e
ex
am
p
l
es
o
f
t
h
e
e
r
r
o
n
e
o
u
s
w
o
r
d
s
e
x
is
t
in
g
in
t
h
e
co
r
p
u
s
w
r
it
te
n
in
L
ati
n
/
T
i
f
i
n
a
g
h
c
h
a
r
ac
te
r
s
,
a
n
d
g
i
v
es t
h
e
co
r
r
ec
t
w
o
r
d
s
i
n
th
e
tw
o
l
ast
c
o
l
u
m
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
ma
z
ig
h
p
a
r
t
-
of
-
s
p
ee
ch
ta
g
g
i
n
g
w
ith
ma
ch
in
e
lea
r
n
i
n
g
a
n
d
d
ee
p
lea
r
n
in
g
(
Otma
n
Ma
a
r
o
u
f)
1817
T
ab
le
1
.
E
x
am
p
le
o
f
th
e
er
r
o
n
eo
u
s
wo
r
d
in
th
e
co
r
p
u
s
W
o
r
d
I
n
C
o
r
p
u
s
Tr
a
n
s
c
o
d
e
d
W
o
r
d
Ti
f
i
n
a
g
h
e
C
o
r
r
e
c
t
W
o
r
d
La
t
i
n
C
o
r
r
e
c
t
W
o
r
d
a
maz
i
G
ⴰ
ⵎ
ⴰ
ⵣⵉ
ⴳ
ⴰ
ⵎ
ⴰ
ⵣⵉ
ⵖ
a
maz
i
Ɣ
nG
ⵏⴳ
ⵏⵖ
nƔ
A
k
k
w
ⴰ
ⴽⴽ
ⵡ
ⴰ
ⴽⴽ
ⵯ
a
k
k
w
t
d
g
g
w
a
t
ⵜ
ⴷⴳⴳⵡⴰ
ⵜ
ⵜ
ⴰ
ⴷⴳⴳ
ⵯ
ⴰ
ⵜ
t
d
g
g
w
a
t
iD
ⵉⴷ
ⵉⴹ
i
ḍ
i
Zl
mD
ⵉ
ⵣⵍ
ⵎ
ⴷ
ⵉ
ⵥ
ⵍⵎⴹ
i
ẓ
lm
ḍ
i
R
a
Za
G
n
ⵉ
ⵔ
ⴰ
ⵣ
ⴰ
ⴳⵏ
ⵉ
ⵕ
ⴰ
ⵥ
ⴰ
ⵥ
ⵖ
ⵏ
i
ṛ
a
ẓ
a
Ɣ
n
l
mas
i
H
ⵍⵎⴰ
ⵙ
ⵉ
ⵀ
ⵍⵎⴰ
ⵙ
ⵉ
ⵃ
l
mas
i
ḥ
mi
S
R
ⵎ
ⵉ
ⵙ
ⵔ
ⵎ
ⵉ
ⵚ
ⵕ
mi
ṣṛ
Ei
c
a
ⴻⵉ
ⵛⴰ
ⵄ
ⵉ
ⵛⴰ
ɛ
i
c
a
i
H
y
a
W
i
ⵉ
ⵀ
ⵢ
ⴰ
ⵡ
ⵉ
ⵉ
ⵃⵢ
ⴰ
ⵡ
ⵉ
i
ḥ
y
a
w
i
u
T
u
ma
t
i
k
ⵓ
ⵜ
ⵓ
ⵎ
ⴰ
ⵜ
ⵉ
ⴽ
ⵡ
ⵟ
ⵓ
ⵎ
ⴰ
ⵜ
ⵉ
ⴽ
u
ṭ
u
ma
t
i
k
4.
AL
G
O
RIT
H
M
S A
ND
RE
S
UL
T
S
4
.
1
.
Dee
p lea
rning
a
lg
o
rit
h
m
s
4
.
1
.
1
.
Rec
urre
nt
neura
l net
wo
rk
(
RNN)
A
r
e
c
u
r
r
e
n
t
n
eu
r
al
n
et
wo
r
k
(
R
NN)
is
a
class
o
f
a
r
t
if
ici
al
n
e
u
r
al
n
etw
o
r
k
s
wh
e
r
e
ass
o
cia
ti
o
n
s
b
et
wee
n
n
o
d
es
s
t
r
u
ct
u
r
e
a
co
o
r
d
i
n
at
ed
d
ia
g
r
a
m
al
o
n
g
a
t
r
an
s
ie
n
t
ar
r
a
n
g
e
m
e
n
t
.
T
h
is
al
lo
ws
it
t
o
s
h
o
w
f
l
ee
t
in
g
u
n
i
q
u
e
co
n
d
u
ct
.
As
o
b
t
ai
n
e
d
f
r
o
m
th
e
f
e
ed
f
o
r
wa
r
d
n
e
u
r
al
n
etw
o
r
k
,
R
NNs c
an
u
s
e
th
ei
r
i
n
n
e
r
s
t
at
e
(
m
e
m
o
r
y
)
t
o
h
a
n
d
le
v
a
r
i
ab
le
l
e
n
g
th
g
r
o
u
p
in
g
s
o
f
i
n
f
o
r
m
ati
o
n
s
o
u
r
ce
s
.
T
h
is
m
ak
es
t
h
e
m
m
at
er
ial
to
d
if
f
e
r
en
t
t
ask
s
,
f
o
r
ex
am
p
l
e,
u
n
s
e
g
m
en
te
d
,
c
o
n
n
e
cte
d
h
an
d
wr
it
in
g
r
ec
o
g
n
iti
o
n
o
r
s
p
ee
c
h
r
ec
o
g
n
i
ti
o
n
[
2
1
]
.
T
h
e
f
o
ll
o
wi
n
g
e
q
u
ati
o
n
s
h
o
ws
th
e
u
n
r
o
ll
e
d
R
NNs
,
t
h
e
y
iel
d
ℎ
at
s
p
e
ci
f
ic
t
im
e
s
t
e
p
'
'
".
ℎ
=
∑
=
0
+
ℎ
−
1
+
(
1
)
ℎ
−
1
i
s
p
r
e
v
i
o
u
s
o
u
t
p
u
t
a
n
d
is
c
u
r
r
en
t
i
n
p
u
t
a
n
d
r
ep
r
es
e
n
t
s
wei
g
h
t
at
t
h
e
ti
m
e
s
t
ep
,
als
o
U
r
ep
r
ese
n
t
s
wei
g
h
t
ass
o
ci
a
te
d
wit
h
o
u
t
p
u
t
ℎ
−
1
an
d
b
r
e
p
r
ese
n
ts
b
ias
te
r
m
s
.
4
.
1
.
2
.
L
o
ng
s
ho
rt
-
t
er
m m
e
mo
ry
net
wo
r
k
s
(
L
ST
M
)
Ho
ch
r
eiter
an
d
Sch
m
id
h
u
b
e
r
[
2
2
]
p
r
o
p
o
s
e
d
th
e
l
o
n
g
s
h
o
r
t
-
te
r
m
m
e
m
o
r
y
c
ell
(
L
S
T
M)
f
o
r
its
f
a
v
o
r
ab
le
p
o
s
it
io
n
i
n
th
e
q
u
ic
k
er
u
n
i
o
n
a
n
d
i
d
e
n
t
if
y
i
n
g
a
n
d
r
e
m
e
m
b
e
r
i
n
g
lo
n
g
-
te
r
m
d
ep
e
n
d
en
ci
es.
No
tw
it
h
s
ta
n
d
i
n
g
th
e
s
h
o
r
t
-
te
r
m
s
tat
e,
t
h
e
L
ST
M
ce
l
l
h
t
h
as
a
l
o
n
g
-
t
er
m
s
ta
te
.
T
h
e
tw
o
s
t
ates
a
r
e
ele
m
e
n
ts
o
f
th
e
cu
r
r
e
n
t
o
u
tp
u
t
an
d
t
h
e
p
as
t
s
ta
tes
as
s
u
m
m
e
d
u
p
in
t
h
e
ac
co
m
p
a
n
y
in
g
C
t
f
u
n
cti
o
n
s
.
T
h
e
ce
ll
y
i
el
d
is
ju
s
t th
e
s
h
o
r
t
-
t
er
m
s
ta
te
.
=
(
+
ℎ
ℎ
−
1
+
−
1
)
,
=
σ
(
+
ℎ
ℎ
−
1
+
−
1
)
,
=
(
+
ℎ
ℎ
−
1
+
−
1
)
,
=
⊙
−
1
+
⊙
ℎ
(
+
ℎ
ℎ
−
1
)
,
ℎ
=
⊙
ℎ
(
)
,
(
2
)
x
i
s
t
h
e
i
n
p
u
t
v
ec
t
o
r
,
h
is
t
h
e
o
u
t
p
u
t
v
ec
t
o
r
an
d
c
ar
e
th
e
ce
ll
s
t
ate
.
T
h
e
s
u
b
s
c
r
i
p
t
t
r
e
p
r
es
en
ts
t
h
e
cu
r
r
e
n
t
ti
m
e
an
d
t
-
1
is
t
h
e
l
as
t
t
im
e.
I
s
a
s
i
g
m
o
id
f
u
n
cti
o
n
,
ʘ
is
t
h
e
H
a
d
a
m
a
r
d
p
r
o
d
u
c
t
a
n
d
W
r
ep
r
es
e
n
ts
u
n
d
ete
r
m
i
n
e
d
p
a
r
a
m
et
er
s
[
2
3
]
.
I
n
(
2
)
,
f
is
t
h
e
f
o
r
g
et
g
a
te
t
h
a
t
d
e
ci
d
es
w
h
a
t
i
n
f
o
r
m
at
i
o
n
n
e
ed
s
t
o
b
e
d
is
c
a
r
d
ed
f
r
o
m
t
h
e
ce
l
l
s
t
ate
.
i
is
t
h
e
in
p
u
t g
a
te
t
h
at
d
e
ci
d
es
w
h
at
i
n
f
o
r
m
at
io
n
n
e
e
d
s
t
o
b
e
s
t
o
r
e
d
i
n
t
h
e
ce
ll
s
t
ate
.
O
I
s
th
e
o
u
t
p
u
t
g
a
te
t
h
at
d
ec
id
es
wh
at
i
n
f
o
r
m
ati
o
n
to
o
u
t
p
u
t
[2
2
]
.
4
.
2
.
T
ra
ini
ng
t
ec
hn
iqu
es
T
r
ain
–
test
s
p
lit:
th
e
p
r
im
ar
y
th
o
u
g
h
t
(
tr
ain
/tes
t
s
p
lit)
is
t
o
d
iv
id
e
th
e
d
ataset
in
to
two
s
ec
tio
n
s
:
a
tr
ain
in
g
s
et,
an
d
a
test
s
et.
A
wh
ile
later
,
u
s
e
th
e
tr
ain
in
g
s
e
t
to
p
r
ep
ar
e
t
h
e
m
o
d
el
[2
4
]
T
h
is
will
p
er
m
it
u
s
to
ascer
tain
th
e
p
r
esen
tatio
n
o
f
th
e
test
s
et,
an
d
th
e
o
u
tco
m
e
will
b
e
a
d
ec
en
t
esti
m
ate
o
f
th
e
ex
h
ib
itio
n
o
n
o
b
s
cu
r
e
i
n
f
o
r
m
atio
n
.
T
h
e
in
f
o
r
m
ati
o
n
al
in
d
ex
is
is
o
lated
in
t
o
a
tr
ain
in
g
s
et
a
n
d
a
test
s
et.
T
h
e
test
s
et
is
n
o
t
u
s
ed
to
tr
ain
th
e
m
o
d
el,
h
o
w
ev
er
,
ju
s
t
to
ass
ess
it.
I
n
o
u
r
ca
s
e
th
e
d
iv
is
io
n
we
m
ak
e
i
s
n
o
t
ar
b
itra
r
y
,
f
o
r
in
s
tan
ce
,
a
s
u
b
s
et
o
f
o
u
r
in
f
o
r
m
atio
n
in
co
r
p
o
r
ated
th
e
lo
n
e
co
n
ten
t
o
f
a
s
p
ec
if
ic
class
.
W
e
tr
ai
n
e
d
all
th
e
alg
o
r
it
h
m
s
o
n
8
0
%
o
f
t
h
e
c
o
r
p
u
s
a
n
d
we
t
este
d
t
h
e
m
b
y
2
0
%
o
f
t
h
e
c
o
r
p
u
s
.
T
h
is
will
r
esu
lt
in
an
"o
v
er
f
itti
n
g
"
o
v
er
f
itti
n
g
is
a
m
ea
s
u
r
ab
le
e
x
am
in
atio
n
th
at
c
o
m
p
ar
es
t
o
i
n
ten
tly
o
r
p
r
ec
is
ely
to
a
s
p
ec
i
f
ic
ar
r
an
g
em
en
t
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
24
,
No
.
3
,
Dec
em
b
er
2
0
2
1
:
1
8
1
4
-
1
8
2
2
1818
in
f
o
r
m
atio
n
.
C
o
n
s
eq
u
e
n
tly
,
th
is
in
v
esti
g
atio
n
m
ay
n
o
t
r
elat
e
to
ex
tr
a
in
f
o
r
m
atio
n
o
r
d
ep
en
d
ab
ly
an
ticip
ate
f
u
tu
r
e
p
er
ce
p
tio
n
s
.
W
e
ar
e
att
em
p
tin
g
to
s
tay
awa
y
f
r
o
m
th
a
t w
ith
cr
o
s
s
-
ap
p
r
o
v
al.
C
r
o
s
s
-
v
alid
atio
n
:
is
a
s
tr
ateg
y
th
at
p
er
m
its
y
o
u
to
u
s
e
th
e
d
atas
et
f
o
r
p
r
ep
ar
atio
n
an
d
v
al
id
atio
n
,
it
f
ills
in
as
f
o
llo
ws:
T
h
e
d
ataset
is
p
ar
ted
in
to
k
p
ar
ts
o
r
o
v
er
l
ap
.
E
v
er
y
o
n
e
o
f
th
e
k
f
o
l
d
s
is
u
s
ed
th
en
as
a
test
s
et.
T
h
e
r
est
(
th
e
a
s
s
o
ciatio
n
o
f
th
e
k
-
1
d
if
f
er
e
n
t
p
ar
ts
)
is
u
s
ed
f
o
r
tr
ain
in
g
.
E
v
en
tu
al
ly
,
ea
ch
p
o
in
t
(
o
r
p
er
ce
p
tio
n
)
is
u
s
ed
o
n
ce
in
a
t
est s
et
(
k
-
1
)
is
u
s
ed
in
a
tr
ai
n
in
g
s
et
[
2
5
]
.
4
.
3
.
P
r
o
po
s
ed
m
o
del f
o
r
A
ma
zig
h pa
rt
o
f
s
peec
h
W
e
h
a
v
e
th
e
p
r
i
v
i
le
g
e
t
o
a
p
p
l
y
t
h
e
d
e
ep
l
ea
r
n
i
n
g
tec
h
n
i
q
u
es
o
n
th
e
Am
az
ig
h
l
an
g
u
a
g
e
to
p
r
e
d
ic
t
p
ar
t
o
f
s
p
ee
c
h
o
f
w
o
r
d
s
w
r
it
te
n
i
n
T
if
in
a
g
h
c
h
a
r
ac
t
er
s
u
s
i
n
g
t
h
e
o
p
e
n
-
s
o
u
r
ce
li
b
r
a
r
y
K
er
as.
I
n
t
h
is
ar
ticle,
we
u
s
ed
two
alg
o
r
ith
m
s
f
o
r
d
ee
p
lear
n
in
g
wh
ic
h
h
a
v
e
t
h
e
s
am
e
la
y
er
s
.
T
h
e
f
ir
s
t
lay
er
is
in
p
u
t
lay
er
,
it
ac
ce
p
ts
v
ec
to
r
s
o
f
s
h
ap
e
(
28
)
a
n
d
m
atc
h
es o
u
r
X
v
ar
iab
le
(
we
h
av
e
2
8
to
k
e
n
s
in
ea
ch
o
f
o
u
r
s
eq
u
en
ce
s
tr
ai
n
an
d
test
)
.
Nex
t,
we
h
av
e
th
e
e
m
b
ed
d
in
g
lay
er
.
T
h
is
lay
er
will
tak
e
ea
ch
o
f
o
u
r
to
k
en
s
/wo
r
d
s
an
d
tu
r
n
it
in
to
a
d
en
s
e
v
e
cto
r
o
f
s
ize
1
0
0
.
T
h
in
k
o
f
it
as
a
g
ian
t
lo
o
k
u
p
tab
le
(
o
r
d
ictio
n
ar
y
)
with
to
k
en
s
as
k
ey
s
an
d
th
e
ac
tu
al
v
ec
to
r
s
as
v
alu
es.
T
h
is
lo
o
k
u
p
tab
le
is
tr
ain
ab
le,
i.e
.
,
ea
ch
ep
o
ch
d
u
r
in
g
th
e
m
o
d
el
tr
ain
in
g
;
we
u
p
d
ate
d
th
o
s
e
v
ec
to
r
s
to
m
atch
o
u
t
th
e
in
p
u
t.
Af
ter
t
he
e
m
b
e
d
d
in
g
la
y
er
,
o
u
r
in
p
u
t
tu
r
n
s
f
r
o
m
a
v
ec
to
r
o
f
len
g
th
2
8
to
a
m
atr
ix
o
f
s
ize
(
2
8
,
1
0
0
)
.
E
ac
h
o
f
th
e
2
8
to
k
e
n
s
n
o
w
h
as a
v
e
cto
r
o
f
s
ize
1
0
0
.
On
ce
we
h
av
e
th
is
,
we
ca
n
u
s
e
th
e
L
STM
lay
er
(
o
r
th
e
s
im
p
le
R
NN
lay
er
)
th
at
f
o
r
ea
ch
to
k
en
will
lo
o
k
b
o
th
way
s
in
th
e
s
en
ten
c
e.
T
h
e
o
u
tp
u
t
o
f
th
is
lay
er
is
a
m
atr
ix
o
f
s
ize
(
2
8
,
6
4
)
.
Nex
t,
we
h
av
e
d
r
o
p
o
u
t
lay
er
th
at
is
a
m
atr
ix
o
f
s
ize
(
2
8
,
6
4
)
.
Fin
ally
,
we
h
av
e
a
t
im
e
d
is
tr
ib
u
ted
d
en
s
e
lay
er
.
I
t
tak
es
th
e
(
2
8
,
6
4
)
m
atr
ix
o
f
th
e
L
STM
(
o
r
s
im
p
l
e
R
NN)
lay
er
o
u
tp
u
t.
T
ab
le
2
an
d
T
ab
le
3
s
h
o
ws o
u
r
m
o
d
el
lay
er
s
.
I
n
th
e
m
o
d
e
l
we
ch
o
s
e
to
u
s
e
B
atch
No
r
m
aliza
tio
n
lay
er
,
th
e
r
ec
tifie
d
lin
e
ar
u
n
it
(
R
eL
U
)
f
u
n
ctio
n
in
th
e
h
id
d
en
lay
e
r
,
an
d
th
e
s
o
f
ttm
ax
f
u
n
ctio
n
in
th
e
o
u
tp
u
t la
y
er
,
we
u
s
ed
Ad
am
m
e
th
o
d
s
as a
n
o
p
tim
izer
.
T
ab
le
2
.
Pro
p
o
s
ed
L
STM
m
o
d
el
ar
ch
itectu
r
e
La
y
e
r
O
u
t
p
u
t
S
h
a
p
e
p
a
r
a
m
e
t
e
r
s
I
n
p
u
t
La
y
e
r
[
(
N
o
n
e
,
2
8
)
]
0
Emb
e
d
d
i
n
g
(
N
o
n
e
,
2
8
,
1
0
0
)
8
0
9
1
0
0
LSTM
(
N
o
n
e
,
6
4
)
4
2
2
4
0
D
r
o
p
o
u
t
(
N
o
n
e
,
2
8
,
6
4
)
0
D
e
n
se
(
N
o
n
e
,
2
8
,
3
2
)
2
0
8
0
T
ab
le
3
.
Pro
p
o
s
ed
R
NN
m
o
d
e
l a
r
ch
itectu
r
e
La
y
e
r
O
u
t
p
u
t
S
h
a
p
e
p
a
r
a
m
e
t
e
r
s
I
n
p
u
t
La
y
e
r
[
(
N
o
n
e
,
2
8
)
]
0
Emb
e
d
d
i
n
g
(
N
o
n
e
,
2
8
,
1
0
0
)
8
0
9
1
0
0
S
i
mp
l
e
R
N
N
(
N
o
n
e
,
6
4
)
1
0
5
6
0
D
r
o
p
o
u
t
(
N
o
n
e
,
2
8
,
6
4
)
0
D
e
n
se
(
N
o
n
e
,
2
8
,
3
2
)
2
0
8
0
T
h
e
em
b
ed
d
i
n
g
lay
er
th
at
r
ec
e
iv
ed
as a
n
in
p
u
t a
v
ec
to
r
o
f
len
g
th
2
8
an
d
tu
r
n
s
to
a
m
atr
ix
o
f
s
ize
(
2
8
,
1
0
0
)
,
h
as
8
0
9
,
1
0
0
p
a
r
am
eter
s
,
th
e
L
STM
lay
er
h
as
4
2
,
2
4
0
,
th
e
last
lay
er
h
as
2
,
0
8
0
p
ar
am
eter
s
.
So
,
th
e
n
u
m
b
er
t
o
tal
o
f
p
ar
am
ete
r
s
is
:
8
5
3
4
2
0
,
th
e
tr
ain
a
b
le
p
ar
a
m
eter
s
:
8
5
3
4
2
0
,
an
d
n
o
n
-
tr
ain
ab
le
p
ar
am
eter
s
:
0
.
T
h
e
e
m
b
ed
d
in
g
lay
e
r
r
ec
eiv
e
d
as
an
in
p
u
t
a
v
ec
to
r
o
f
len
g
th
2
8
an
d
tu
r
n
s
to
a
m
atr
ix
o
f
s
ize
(
2
8
,
1
0
0
)
h
as
8
0
9
,
1
0
0
p
ar
am
ete
r
s
,
th
e
s
im
p
le
R
NN
lay
er
h
as
1
0
,
5
6
0
an
d
th
e
last
lay
er
h
as
2
,
0
8
0
p
a
r
am
eter
s
.
So
,
th
e
to
tal
n
u
m
b
er
o
f
p
ar
am
eter
s
is
:
8
2
1
,
7
4
0
,
th
e
tr
ain
ab
le
p
ar
am
eter
s
:
821,
7
4
0
,
a
n
d
n
o
n
-
tr
ain
ab
le
p
a
r
am
eter
s
: 0
.
4
.
4
.
Dis
cus
s
io
n
I
n
t
h
is
s
t
u
d
y
,
w
e
f
i
x
e
d
t
h
e
s
iz
e
ce
lls
o
f
R
NN
l
ay
er
in
6
4
,
a
n
d
w
e
v
a
r
i
ed
t
h
e
b
at
ch
s
i
ze
b
etw
e
en
3
2
a
n
d
1
2
8
.
W
e
ch
o
s
e
th
e
n
u
m
b
er
o
f
ep
o
c
h
s
w
h
e
n
t
h
e
ac
c
u
r
ac
y
s
t
o
p
p
e
d
im
p
r
o
v
i
n
g
o
n
t
h
e
tes
t
s
et
s
.
T
r
ai
n
i
n
g
u
s
u
al
ly
s
to
p
s
af
te
r
1
0
e
p
o
c
h
s
.
T
h
e
T
a
b
l
e
4
c
o
n
tai
n
s
t
h
e
r
esu
lts
o
f
t
h
e
d
i
f
f
er
e
n
t
s
ize
b
at
c
h
u
s
i
n
g
R
NN
,
th
ese
r
es
u
lts
s
h
o
w
t
h
at
t
h
e
ac
c
u
r
a
cy
u
s
i
n
g
r
ec
u
r
r
e
n
t
n
e
u
r
a
l n
e
tw
o
r
k
s
(
R
NN
)
g
a
v
e
9
5
.
1
1
%
w
h
e
n
w
e
c
h
o
s
e
3
2
i
n
t
h
e
s
i
z
e
b
atc
h
an
d
9
5
.
2
0
%
in
6
4
a
n
d
9
5
.
0
5
%
i
n
1
2
8
.
T
h
e
r
e
f
o
r
e,
t
h
e
b
est
s
i
ze
b
a
tc
h
is
6
4
.
T
h
e
Fi
g
u
r
e
1
i
n
d
ic
ates
th
e
t
r
ai
n
i
n
g
a
n
d
t
esti
n
g
ac
c
u
r
a
cy
o
f
r
ec
u
r
r
e
n
t
n
e
u
r
al
n
et
wo
r
k
a
lg
o
r
it
h
m
u
s
i
n
g
6
4
s
ize
ce
l
ls
a
n
d
6
4
-
s
i
ze
b
a
tc
h
in
1
0
e
p
o
c
h
s
.
W
e
o
b
s
e
r
v
ed
t
h
at
th
e
a
v
e
r
a
g
e
l
o
s
s
in
t
h
e
las
t
ep
o
c
h
s
is
4
.
7
9
%
.
Fo
r
t
h
e
L
ST
M
n
etw
o
r
k
s
,
we
f
ix
e
d
t
h
e
s
i
ze
c
ells
i
n
6
4
,
a
n
d
we
v
ar
i
e
d
t
h
e
b
atch
s
ize
b
etwe
en
3
2
an
d
128
in
1
0
e
p
o
c
h
s
.
T
h
e
T
a
b
le
5
c
o
n
t
ai
n
s
th
e
r
es
u
lts
o
f
th
e
d
if
f
er
en
t
s
iz
e
b
atc
h
u
s
i
n
g
L
ST
M,
th
es
e
r
es
u
lts
s
h
o
w
th
a
t
t
h
e
b
est
a
cc
u
r
ac
y
,
u
s
i
n
g
L
o
n
g
s
h
o
r
t
-
t
e
r
m
Me
m
o
r
y
n
et
wo
r
k
s
(
L
S
T
M
)
is
9
7
.
8
8
%
i
n
b
atc
h
s
ize
1
2
8
,
w
h
e
r
e
th
e
a
cc
u
r
a
cy
g
i
v
e
n
is
9
7
.
8
1
%
b
y
3
2
-
s
i
ze
b
at
c
h
an
d
6
4
-
s
iz
e
b
atc
h
.
T
h
er
ef
o
r
e
,
th
e
b
est s
iz
e
b
atc
h
is
1
2
8
.
T
ab
le
4
.
POS tag
g
in
g
ac
cu
r
ac
y
u
s
in
g
R
NN
S
i
z
e
b
a
t
c
h
R
N
N
a
c
c
u
r
a
c
y
32
9
5
.
1
1
%
64
9
5
.
2
0
%
1
2
8
9
5
.
0
5
%
T
ab
le
5
.
POS tag
g
in
g
ac
cu
r
ac
y
u
s
in
g
L
STM
S
i
z
e
b
a
t
c
h
LS
T
M
a
c
c
u
r
a
c
y
32
9
7
.
8
1
%
64
9
7
.
8
1
%
1
2
8
9
7
.
8
8
%
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
ma
z
ig
h
p
a
r
t
-
of
-
s
p
ee
ch
ta
g
g
i
n
g
w
ith
ma
ch
in
e
lea
r
n
i
n
g
a
n
d
d
ee
p
lea
r
n
in
g
(
Otma
n
Ma
a
r
o
u
f)
1819
T
h
e
F
ig
u
r
e
2
in
d
icate
s
th
e
tr
ain
in
g
an
d
test
in
g
ac
cu
r
ac
y
o
f
L
o
n
g
Sh
o
r
t
-
ter
m
m
em
o
r
y
alg
o
r
ith
m
u
s
in
g
6
4
s
ize
ce
lls
an
d
1
2
8
-
s
ize
b
atch
in
1
0
ep
o
c
h
s
(
F
ig
u
r
e
2
)
.
W
e
o
b
s
er
v
ed
th
at
th
e
av
e
r
ag
e
lo
s
s
in
th
e
last
ep
o
ch
s
is
2
.
1
1
%.
T
h
e
Fig
u
r
e
3
in
d
icate
s
th
e
d
if
f
er
en
t
v
ar
iat
io
n
s
o
f
th
e
Pre
cisi
o
n
,
R
ec
all
an
d
F1
-
s
co
r
e
in
th
e
two
p
r
o
p
o
s
ed
m
o
d
els (
R
NN
an
d
L
STM
)
,
f
o
r
th
e
p
a
r
t
o
f
s
p
e
ec
h
tag
s
et.
W
e
o
b
s
er
v
e
d
th
at
th
e
m
ajo
r
ity
o
f
POS
tag
h
as
th
e
s
im
ilar
v
alu
es
in
p
r
ec
is
io
n
,
r
ec
all
,
F1
-
s
co
r
e
,
ex
ce
p
t
p
ar
ticle
(
PR
OT
)
r
esid
u
al
(
R
OT
)
,
an
d
d
em
o
n
s
tr
ativ
e
p
r
o
n
o
u
n
(
PDEM
)
,
wh
ich
h
a
v
e
lo
w
v
alu
es,
t
h
at
is
ca
u
s
ed
b
y
th
e
m
is
s
in
g
wo
r
d
s
o
f
th
ese
ty
p
es.
Fig
u
r
e
1
.
R
NN
tr
ain
in
g
a
n
d
te
s
tin
g
ac
cu
r
ac
y
u
s
in
g
6
4
s
ize
ce
lls
an
d
s
ize
b
atch
in
1
0
ep
o
ch
s
Fig
u
r
e
2
.
L
STM
tr
ain
i
n
g
an
d
t
esti
n
g
ac
cu
r
ac
y
u
s
in
g
6
4
s
ize
ce
lls
an
d
1
2
8
s
ize
b
atc
h
in
1
0
ep
o
c
h
s
F
ig
u
r
e
3
.
Pre
cisi
o
n
,
re
ca
ll,
F1
-
s
co
r
e
f
o
r
R
NN
an
d
L
STM
m
o
d
els
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
24
,
No
.
3
,
Dec
em
b
er
2
0
2
1
:
1
8
1
4
-
1
8
2
2
1820
T
o
b
e
o
b
jectiv
e,
we
co
m
p
ar
e
d
o
u
r
L
STM
m
o
d
el
an
d
R
NN
m
o
d
el
to
o
th
er
m
o
d
el
wh
i
ch
u
s
ed
th
e
s
am
e
d
ataset
in
th
e
s
am
e
co
n
d
itio
n
s
,
f
o
r
th
e
ep
o
c
h
n
u
m
b
e
r
we
u
s
ed
;
1
0
e
p
o
ch
s
f
o
r
all
s
y
s
tem
s
.
T
h
e
tab
le
b
elo
w
s
u
m
m
ar
izes
th
e
o
b
tain
ed
r
esu
lts
.
As
s
h
o
wn
in
T
ab
le
6
,
in
t
er
m
s
o
f
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
o
r
f
1
-
s
co
r
e,
o
u
r
L
STM
g
iv
es
b
etter
r
esu
lts
co
m
p
a
r
ed
t
o
o
t
h
er
s
y
s
tem
s
9
7
.
8
8
%
in
f
1
-
s
co
r
e.
T
h
i
s
in
d
icate
s
th
at
o
u
r
s
y
s
tem
is
a
v
er
y
p
o
wer
f
u
l
cla
s
s
if
ier
in
th
e
Am
az
ig
h
p
a
r
t
o
f
s
p
ee
ch
.
T
h
e
a
r
ch
itectu
r
e
p
r
o
p
o
s
ed
in
[
1
6
]
clo
s
er
to
th
is
m
o
d
el
in
ter
m
s
o
f
ac
cu
r
ac
y
g
av
e
9
7
.
8
7
%.
B
u
t
th
e
d
if
f
er
en
ce
in
th
e
n
u
m
b
er
o
f
lay
er
s
u
s
ed
in
th
e
two
m
o
d
els.
T
h
e
m
o
d
el
p
r
o
p
o
s
ed
in
[
1
6
]
u
s
ed
6
lay
e
r
s
an
d
o
u
r
m
o
d
el
u
s
ed
ju
s
t
5
lay
er
s
an
d
is
u
s
ed
in
th
e
h
y
p
er
b
o
lic
tan
g
en
t
as
an
ac
tiv
atio
n
f
u
n
ctio
n
in
th
e
in
p
u
t
lay
er
in
o
u
r
m
o
d
el
we
u
s
ed
R
e
L
U
f
u
n
ctio
n
.
I
n
th
e
o
u
tp
u
t
la
y
er
,
we
u
s
ed
t
h
e
s
o
f
tm
ax
f
u
n
ctio
n
,
th
e
m
o
d
el
p
r
o
p
o
s
ed
in
[
1
6
]
u
s
ed
th
e
s
ig
m
o
id
f
u
n
ctio
n
,
as
a
n
o
p
tim
izer
we
u
s
ed
th
e
Ad
am
o
p
tim
izer
in
th
e
two
m
o
d
els.
T
ab
le
6
.
C
o
m
p
a
r
is
o
n
o
f
o
u
r
L
STM
an
d
R
NN
m
o
d
els with
o
th
er
m
eth
o
d
s
(
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
f1
-
s
co
r
e)
M
e
t
h
o
d
s
A
c
c
u
r
a
c
y
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
D
e
c
i
s
i
o
n
Tr
e
e
[
1
6
]
0
.
8
7
6
0
.
8
7
6
0
.
8
4
2
0
.
8
3
2
CRF
[
1
6
]
0
.
9
3
4
0
.
9
3
4
0
.
9
3
5
0
.
9
3
3
LSTM
[
1
6
]
0
.
9
7
8
7
0
.
9
7
8
6
0
.
9
7
8
7
0
.
9
7
8
7
O
u
r
R
N
N
m
o
d
e
l
0
.
9
5
2
0
0
.
9
5
2
0
0
.
9
5
2
1
0
.
9
5
2
0
O
u
r
LST
M
m
o
d
e
l
0
.
9
7
8
8
0
.
9
7
8
9
0
.
9
7
8
9
0
.
9
7
8
8
R
eg
ar
d
in
g
t
h
e
e
x
ec
u
tio
n
tim
e
d
u
r
in
g
tr
ain
in
g
i
n
T
a
b
le
7
,
o
u
r
s
y
s
tem
s
h
av
e
s
h
o
wn
th
eir
e
f
f
i
cien
cy
.
I
n
th
e
t
r
ain
in
g
p
h
ase,
o
u
r
R
NN
s
y
s
tem
v
er
y
f
ast
co
m
p
ar
ed
t
o
L
STM
m
o
d
els:
4
4
.
3
6
%
co
m
p
ar
ed
to
th
e
p
r
o
p
o
s
ed
L
STM
in
[
1
6
]
an
d
3
0
.
4
0
%
c
o
m
p
ar
ed
t
o
o
u
r
L
STM
m
o
d
e
l,
an
d
o
u
r
L
STM
m
o
d
el
is
v
er
y
f
ast
b
y
2
0
.
0
5
%
co
m
p
ar
ed
to
t
h
e
L
STM
p
r
o
p
o
s
ed
in
[
1
6
]
.
T
h
is
b
ig
d
i
f
f
er
e
n
ce
in
e
x
ec
u
tio
n
tim
e
is
ex
p
lain
ed
b
y
th
e
lar
g
e
n
u
m
b
er
o
f
p
a
r
am
eter
s
u
s
ed
i
n
L
STM
ar
c
h
itectu
r
e
[
1
6
]
w
h
ich
u
s
ed
2
,
5
2
3
,
7
7
7
p
a
r
am
et
er
s
co
m
p
ar
e
d
to
o
u
r
m
o
d
els.
I
n
t
h
e
R
NN
,
we
u
s
ed
8
2
1
,
7
4
0
p
ar
am
eter
s
an
d
i
n
th
e
L
STM
m
o
d
el,
we
u
s
ed
8
5
3
,
4
2
0
p
ar
a
m
eter
s
.
T
ab
le
7
.
C
o
m
p
a
r
is
o
n
o
f
o
u
r
L
STM
an
d
R
NN
m
o
d
els with
o
th
er
m
eth
o
d
s
(
n
u
m
b
er
o
f
p
a
r
a
m
eter
s
,
tr
ain
in
g
tim
e)
M
e
t
h
o
d
s
P
a
r
a
me
t
e
r
s
Tr
a
i
n
i
n
g
t
i
me
(
s)
LSTM
[1
6
]
2
,
5
2
3
,
7
7
7
6
5
.
3
2
O
u
r
R
N
N
m
o
d
e
l
8
2
1
,
7
4
0
3
6
.
3
4
O
u
r
LST
M
m
o
d
e
l
8
5
3
,
4
2
0
5
2
.
2
2
5.
CO
NCLU
SI
O
N
Am
a
zi
g
h
p
a
r
t
o
f
s
p
e
ec
h
is
o
n
e
o
f
t
h
e
m
o
s
t
i
n
te
r
es
ti
n
g
s
u
b
je
cts
t
r
e
ate
d
b
y
s
e
v
er
al
r
ese
ar
ch
er
s
,
i
n
p
a
r
ti
c
u
la
r
M
o
r
o
cc
a
n
r
ese
a
r
c
h
e
r
s
.
I
n
t
h
is
a
r
ti
cle
,
we
u
s
e
d
d
e
ep
le
ar
n
i
n
g
a
lg
o
r
it
h
m
s
t
o
l
a
b
el
w
o
r
d
s
wr
itt
en
i
n
T
if
in
a
g
h
ch
a
r
ac
te
r
s
.
W
e
c
o
m
p
a
r
e
d
o
u
r
R
NN
an
d
L
ST
M
m
o
d
els
wit
h
ex
is
ti
n
g
d
ee
p
le
ar
n
i
n
g
m
et
h
o
d
s
a
n
d
m
ac
h
i
n
e
le
ar
n
i
n
g
m
et
h
o
d
s
,
w
e
f
o
u
n
d
t
h
at
o
u
r
R
N
N
m
o
d
el
is
b
et
t
e
r
t
h
a
n
C
R
F
an
d
d
ec
i
s
io
n
tr
ee
(
m
ac
h
i
n
e
lea
r
n
i
n
g
a
lg
o
r
it
h
m
s
)
t
h
a
t i
t
ap
p
lie
d
in
t
h
e
s
a
m
e
d
at
aset
a
n
d
i
n
th
e
s
a
m
e
c
o
n
d
iti
o
n
s
.
B
y
c
o
m
p
ar
i
n
g
t
h
e
d
ee
p
le
a
r
n
i
n
g
al
g
o
r
it
h
m
s
w
it
h
e
ac
h
o
t
h
e
r
,
w
e
s
e
e
t
h
at
o
u
r
r
ec
u
r
r
e
n
t
n
eu
r
al
n
etw
o
r
k
m
o
d
el
is
wea
k
c
o
m
p
ar
e
d
to
t
h
e
o
t
h
e
r
L
o
n
g
s
h
o
r
t
-
t
er
m
m
em
o
r
y
L
STM
m
o
d
el
i
n
te
r
m
s
o
f
a
cc
u
r
ac
y
b
u
t
ti
m
e
o
f
ex
ec
u
ti
o
n
is
t
h
e
b
est
o
n
e
.
T
h
is
is
b
ec
a
u
s
e
t
h
e
g
r
a
d
ie
n
t
o
f
t
h
e
lo
s
s
f
u
n
c
ti
o
n
d
e
ca
y
s
e
x
p
o
n
en
ti
all
y
wit
h
ti
m
e
(
t
h
e
v
a
n
is
h
i
n
g
g
r
a
d
i
en
t
p
r
o
b
l
e
m
)
.
T
h
is
s
h
o
ws
t
h
at
L
STM
n
et
wo
r
k
s
ca
n
ca
p
t
u
r
e
k
n
o
wl
e
d
g
e
o
f
a
l
a
n
g
u
a
g
e
b
ec
a
u
s
e
L
STM
u
n
its
i
n
cl
u
d
e
a
'
m
e
m
o
r
y
c
ell'
t
h
at
ca
n
m
ai
n
t
ai
n
i
n
f
o
r
m
at
io
n
i
n
m
e
m
o
r
y
f
o
r
l
o
n
g
p
e
r
i
o
d
s
.
A
s
et
o
f
g
ates
is
u
s
ed
t
o
co
n
t
r
o
l
wh
en
i
n
f
o
r
m
at
io
n
e
n
te
r
s
t
h
e
m
e
m
o
r
y
w
h
e
n
it
is
o
u
t
p
u
t
,
an
d
wh
en
it
is
f
o
r
g
o
tte
n
.
T
h
is
ar
ch
ite
ct
u
r
e
lets
t
h
e
m
le
ar
n
l
o
n
g
e
r
-
te
r
m
d
e
p
e
n
d
e
n
c
ies
,
w
h
en
we
c
o
m
p
a
r
e
d
o
u
r
L
STM
m
o
d
el
an
d
t
h
e
t
h
e
o
t
h
e
r
L
STM
m
o
d
el,
w
e
f
o
u
n
d
t
h
a
t o
u
r
m
o
d
el
is
b
ett
er
i
n
te
r
m
s
th
e
ac
c
u
r
a
cy
a
n
d
i
n
te
r
m
s
o
f
e
x
ec
u
ti
o
n
ti
m
e
.
T
h
is
h
as
a
t
r
e
m
e
n
d
o
u
s
p
r
e
f
e
r
r
e
d
p
o
s
iti
o
n
,
p
ar
t
ic
u
l
ar
ly
f
o
r
b
ei
n
g
u
s
e
d
o
n
d
ia
lec
ts
o
n
w
h
ic
h
we
d
o
n
o
t
h
a
v
e
m
a
n
y
lan
g
u
a
g
e
s
t
u
d
ies
.
As
a
p
e
r
s
p
e
cti
v
e
,
we
im
p
r
o
v
ed
th
e
o
b
tai
n
ed
r
es
u
l
ts
b
y
p
r
o
p
o
s
i
n
g
o
t
h
er
ap
p
r
o
ac
h
es
to
la
b
e
l
wo
r
d
s
w
r
it
te
n
in
T
if
in
a
g
h
c
h
ar
ac
t
er
s
a
n
d
tr
ea
t
o
t
h
e
r
tas
k
s
in
t
h
e
s
a
m
e
d
o
m
a
in
.
RE
F
E
R
E
NC
E
S
[1
]
M
.
Bo
u
k
a
b
o
u
s
a
n
d
M
.
Az
izi,
“
A
c
o
m
p
a
ra
ti
v
e
stu
d
y
o
f
d
e
e
p
l
e
a
rn
in
g
b
a
se
d
lan
g
u
a
g
e
re
p
re
se
n
tatio
n
lea
rn
in
g
m
o
d
e
ls
,”
In
d
o
n
e
sia
n
J
o
u
rn
a
l
o
f
E
lec
trica
l
En
g
i
n
e
e
rin
g
a
n
d
Co
m
p
u
ter
S
c
ien
c
e
,
v
o
l.
2
2
,
n
o
.
2
,
p
p
.
1
0
3
2
-
1
0
4
0
,
2
0
2
1
,
d
o
i:
1
0
.
1
1
5
9
1
/
ij
e
e
c
s.v
2
2
.
i2
.
p
p
1
0
3
2
-
1
0
4
0
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
ma
z
ig
h
p
a
r
t
-
of
-
s
p
ee
ch
ta
g
g
i
n
g
w
ith
ma
ch
in
e
lea
r
n
i
n
g
a
n
d
d
ee
p
lea
r
n
in
g
(
Otma
n
Ma
a
r
o
u
f)
1821
[2
]
D.
Ju
ra
fsk
y
a
n
d
J.
H.
M
a
rti
n
,
S
p
e
e
c
h
a
n
d
l
a
n
g
u
a
g
e
p
ro
c
e
ss
in
g
:
a
n
i
n
tro
d
u
c
ti
o
n
t
o
n
a
t
u
ra
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ss
in
g
,
c
o
mp
u
t
a
ti
o
n
a
l
li
n
g
u
isti
c
s,
a
n
d
sp
e
e
c
h
re
c
o
g
n
it
i
o
n
.
Ne
w Je
rse
y
:
USA:
P
re
n
ti
c
e
Ha
ll
,
2
0
0
0
.
[3
]
V.
G
u
p
ta,
N.
Jo
sh
i,
a
n
d
I.
M
a
th
u
r,
“
CRF
b
a
se
d
P
a
rt
o
f
S
p
e
e
c
h
Ta
g
g
e
r
fo
r
Do
m
a
in
S
p
e
c
ifi
c
Hin
d
i
Co
rp
u
s
,”
IJ
CA
Pro
c
e
e
d
in
g
s
o
n
Na
ti
o
n
a
l
C
o
n
f
.
o
n
Co
n
tem
p
o
r
a
ry
Co
mp
u
ti
n
g
,
n
o
.
2
,
2
0
1
7
,
p
p
.
1
4
-
1
8
.
[4
]
T.
Ta
n
,
B.
Ra
n
a
iv
o
-
M
a
lan
ç
o
n
,
L
.
Be
sa
c
ier,
Y.
Ye
o
n
g
,
K.
H.
G
a
n
,
a
n
d
E
.
K.
Ta
n
g
,
“
Ev
a
l
u
a
ti
n
g
LS
TM
Ne
two
r
k
s,
HMM
a
n
d
W
F
S
T
i
n
M
a
lay
P
a
rt
-
of
-
S
p
e
e
c
h
Tag
g
in
g
,”
J
o
u
rn
a
l
o
f
T
e
lec
o
mm
u
n
ica
ti
o
n
,
El
e
c
tro
n
ic
a
n
d
C
o
mp
u
ter
En
g
i
n
e
e
rin
g
(J
T
EC)
,
v
o
l
.
9
,
n
o
.
2
-
9
,
p
p
.
7
9
-
8
3
,
2
0
1
7
.
[5
]
K.
Ku
m
a
r
a
n
d
D.
P
.
G
a
n
d
h
m
a
l,
“
An
in
telli
g
e
n
t
i
n
d
ian
sto
c
k
m
a
rk
e
t
fo
re
c
a
stin
g
sy
ste
m
u
si
n
g
L
S
TM
d
e
e
p
lea
rn
in
g
,”
In
d
o
n
e
sia
n
J
o
u
rn
a
l
o
f
E
lec
trica
l
En
g
in
e
e
rin
g
a
n
d
C
o
mp
u
ter
S
c
ien
c
e
,
v
o
l
.
2
1
,
n
o
.
2
,
p
p
.
1
0
8
2
-
1
0
8
9
,
2
0
2
1
,
d
o
i:
1
0
.
1
1
5
9
1
/
ij
e
e
c
s.v
2
1
.
i2
.
p
p
1
0
8
2
-
1
0
8
9
.
[6
]
A.
M
.
S
.
Om
a
r,
M
.
K.
Os
m
a
n
,
M
.
N.
Ib
ra
h
im,
Z.
Hu
ss
a
in
,
a
n
d
A.
F
.
Ab
id
i
n
,
“
F
a
u
lt
c
l
a
ss
ifi
c
a
ti
o
n
o
n
tran
sm
issio
n
li
n
e
u
si
n
g
LS
TM
n
e
two
r
k
,”
In
d
o
n
e
sia
n
J
o
u
rn
a
l
o
f
El
e
c
trica
l
En
g
in
e
e
rin
g
a
n
d
Co
m
p
u
ter
S
c
ien
c
e
,
v
o
l.
2
0
,
n
o
.
1
,
p
p
.
2
3
1
-
238
,
2
0
2
0
,
d
o
i:
1
0
.
1
1
5
9
1
/i
jee
c
s.v
2
0
.
i1
.
p
p
2
3
1
-
2
3
8
.
[7
]
K.
To
u
tan
o
v
a
,
D.
Kle
in
,
C.
D.
M
a
n
n
i
n
g
,
a
n
d
Y
.
S
i
n
g
e
r,
“
F
e
a
tu
re
-
rich
p
a
rt
-
of
-
sp
e
e
c
h
tag
g
in
g
with
a
c
y
c
li
c
d
e
p
e
n
d
e
n
c
y
n
e
tw
o
rk
,”
Pr
o
c
e
e
d
in
g
s
o
f
t
h
e
2
0
0
3
Co
n
fer
e
n
c
e
o
f
t
h
e
No
rth
Ame
ric
a
n
Ch
a
p
ter
o
f
th
e
Asso
c
ia
ti
o
n
f
o
r
Co
mp
u
t
a
ti
o
n
a
l
L
in
g
u
isti
c
s
o
n
Hu
ma
n
L
a
n
g
u
a
g
e
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
1
,
2
0
0
3
,
p
p
.
1
7
3
-
1
8
0
.
d
o
i:
1
0
.
3
1
1
5
/
1
0
7
3
4
4
5
.
1
0
7
3
4
7
8
.
[8
]
J.
G
im
é
n
e
z
a
n
d
L.
M
à
rq
u
e
z
,
“
S
VMT
o
o
l:
A
g
e
n
e
ra
l
P
OS
Tag
g
e
r
G
e
n
e
ra
to
r
Ba
se
d
o
n
S
u
p
p
o
rt
Ve
c
to
r
M
a
c
h
in
e
s
,”
in
Pro
c
e
e
d
in
g
s
o
f
t
h
e
F
o
u
rt
h
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
L
a
n
g
u
a
g
e
Res
o
u
rc
e
s
a
n
d
Eva
lu
a
ti
o
n
(L
RE
C’0
4
)
,
2
0
0
4
,
p
p
.
4
3
–
6
1
.
[9
]
M
.
Bi
n
iz
a
n
d
R.
El
Ay
a
c
h
i
,
“
Re
c
o
g
n
it
io
n
o
f
Ti
fi
n
a
g
h
Ch
a
ra
c
ters
Us
in
g
O
p
ti
m
ize
d
C
o
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
,”
S
e
n
s Ima
g
in
g
,
v
o
l
.
2
2
,
n
o
.
2
8
,
2
0
2
1
,
d
o
i:
1
0
.
1
0
0
7
/s
1
1
2
2
0
-
0
2
1
-
0
0
3
4
7
-
1.
[1
0
]
J.
D.
Laffe
rty
,
A.
M
c
Ca
ll
u
m
,
a
n
d
F
.
C.
N.
P
e
re
ira,
“
Co
n
d
it
i
o
n
a
l
Ra
n
d
o
m
F
ield
s:
P
r
o
b
a
b
il
is
ti
c
M
o
d
e
ls
fo
r
S
e
g
m
e
n
ti
n
g
a
n
d
Lab
e
li
n
g
S
e
q
u
e
n
c
e
Da
ta
,”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
E
ig
h
tee
n
t
h
I
n
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
M
a
c
h
in
e
L
e
a
rn
in
g
,
2
0
0
1
,
p
p
.
2
8
2
-
2
8
9
.
[1
1
]
Y.
Tsu
r
u
o
k
a
,
J.
Ts
u
ji
i,
a
n
d
S
.
An
a
n
iad
o
u
,
“
F
a
st
F
u
ll
P
a
rsin
g
b
y
Li
n
e
a
r
-
Ch
a
in
Co
n
d
it
io
n
a
l
Ra
n
d
o
m
F
ield
s
,”
in
Pro
c
e
e
d
in
g
s
o
f
th
e
1
2
th
Co
n
fer
e
n
c
e
o
f
t
h
e
E
u
ro
p
e
a
n
Ch
a
p
ter
o
f
t
h
e
ACL
(EA
C
L
2
0
0
9
)
,
2
0
0
9
,
p
p
.
7
9
0
-
7
9
8
,
d
o
i:
1
0
.
3
1
1
5
/1
6
0
9
0
6
7
.
1
6
0
9
1
5
5
.
[1
2
]
T.
Bra
n
ts,
“
T
n
T:
A
S
tatisti
c
a
l
P
a
rt
-
of
-
sp
e
e
c
h
Tag
g
e
r
,”
S
ixt
h
A
p
p
li
e
d
Na
t
u
ra
l
L
a
n
g
u
a
g
e
Pr
o
c
e
ss
in
g
C
o
n
fer
e
n
c
e
,
2
0
0
0
,
p
p
.
2
2
4
–
2
3
1
.
d
o
i:
1
0
.
3
1
1
5
/
9
7
4
1
4
7
.
9
7
4
1
7
8
.
[1
3
]
M
.
Ou
tah
a
jala
,
Y.
Be
n
a
ji
b
a
,
P
.
Ro
ss
o
,
a
n
d
L.
Zen
k
o
u
a
r,
“
P
O
S
Tag
g
in
g
i
n
Am
a
z
ig
h
e
Us
in
g
S
u
p
p
o
rt
Ve
c
to
r
M
a
c
h
in
e
s
a
n
d
C
o
n
d
it
i
o
n
a
l
Ra
n
d
o
m
F
iel
d
s
,”
I
n
ter
n
a
ti
o
n
a
l
Co
n
fe
re
n
c
e
o
n
Ap
p
li
c
a
ti
o
n
o
f
Na
t
u
ra
l
L
a
n
g
u
a
g
e
to
In
fo
rm
a
t
io
n
S
y
ste
ms
,
2
0
1
1
,
p
p
.
2
3
8
-
2
4
1
,
d
o
i:
1
0
.
1
0
0
7
/9
7
8
-
3
-
6
4
2
-
2
2
3
2
7
-
3
_
2
8
.
[1
4
]
S
.
Am
ri
a
n
d
L.
Zen
k
o
u
a
r,
“
Am
a
z
ig
h
P
OS
Tag
g
in
g
Us
in
g
T
re
e
Tag
g
e
r:
A
Lan
g
u
a
g
e
I
n
d
e
p
e
n
d
a
n
t
M
o
d
e
l
,”
In
ter
n
a
t
io
n
a
l
Co
n
f
e
re
n
c
e
o
n
Ad
v
a
n
c
e
d
I
n
telli
g
e
n
t
S
y
ste
ms
f
o
r
S
u
sta
in
a
b
le
De
v
e
lo
p
me
n
t
,
p
p
.
6
2
2
-
6
3
2
,
2
0
1
9
,
d
o
i:
1
0
.
1
0
0
7
/9
7
8
-
3
-
0
3
0
-
1
1
9
2
8
-
7
_
5
6
.
[1
5
]
S
.
Am
ri,
L
.
Zen
k
o
u
a
r,
a
n
d
R
.
Be
n
k
h
o
u
y
a
,
‘Am
a
z
ig
h
e
P
OS
tag
g
i
n
g
u
si
n
g
Lo
n
g
S
h
o
rt
Term
M
e
m
o
r
y
Ne
two
rk
s’
,
i
n
Pro
c
e
e
d
in
g
s
o
f
t
h
e
4
t
h
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
B
ig
Da
ta
a
n
d
I
n
ter
n
e
t
o
f
T
h
i
n
g
s
,
n
o
.
4
3
,
2
0
1
9
,
p
p
.
1
-
5
,
d
o
i:
1
0
.
1
1
4
5
/3
3
7
2
9
3
8
.
3
3
7
2
9
8
1
.
[1
6
]
O.
M
a
a
ro
u
f
a
n
d
R.
El
Ay
a
c
h
i,
“
P
a
rt
-
of
-
S
p
e
e
c
h
Tag
g
i
n
g
Us
in
g
L
o
n
g
S
h
o
rt
Term
M
e
m
o
ry
(LS
T
M
):
Am
a
z
ig
h
Tex
t
Wr
it
ten
in
Ti
f
in
a
g
h
e
Ch
a
ra
c
ters
,”
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
Bu
si
n
e
ss
In
telli
g
e
n
c
e
,
p
p
.
3
-
17
,
2
0
2
1
,
d
o
i:
1
0
.
1
0
0
7
/9
7
8
-
3
-
0
3
0
-
7
6
5
0
8
-
8
_
1
.
[1
7
]
A.
S
a
m
ir,
Z.
Lah
b
i
b
,
a
n
d
O.
M
o
h
a
m
e
d
,
“
Am
a
z
ig
h
P
o
S
Ta
g
g
i
n
g
Us
in
g
M
a
c
h
i
n
e
Lea
rn
in
g
Tec
h
n
i
q
u
e
s
,”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
M
e
d
it
e
rr
a
n
e
a
n
S
y
mp
o
si
u
m
o
n
S
ma
rt
Cit
y
Ap
p
l
ica
ti
o
n
s
,
p
p
.
5
5
1
-
5
6
2
,
2
0
1
8
,
d
o
i:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
319
-
7
4
5
0
0
-
8
_
5
1
.
[1
8
]
S
.
Am
ri,
L.
Zen
k
o
u
a
r,
a
n
d
M
.
O
u
tah
a
jala
,
“
Am
a
z
ig
h
P
a
rt
-
of
-
S
p
e
e
c
h
Tag
g
in
g
Us
i
n
g
M
a
rk
o
v
M
o
d
e
l
s
a
n
d
De
c
isio
n
Tree
s
,”
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
Co
mp
u
ter
S
c
ien
c
e
&
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
lo
g
y
,
v
o
l
.
8
,
n
o
.
5
,
p
p
.
6
1
-
7
1
,
2
0
1
6
,
d
o
i:
1
0
.
5
1
2
1
/i
jcs
it
.
2
0
1
6
.
8
5
0
5
.
[1
9
]
M
.
Ou
tah
a
jala
,
L
.
Ze
k
o
u
a
r,
P
.
R
o
ss
o
,
a
n
d
M
.
M
.
An
t
ò
n
ia,
“
Tag
g
in
g
Am
a
z
ig
h
with
An
C
o
ra
P
i
p
e
,”
Pr
o
c
e
e
d
in
g
o
f
t
h
e
W
o
rk
sh
o
p
o
n
L
a
n
g
u
a
g
e
Res
o
u
rc
e
s a
n
d
Hu
ma
n
L
a
n
g
u
a
g
e
T
e
c
h
n
o
lo
g
y
fo
r
S
e
miti
c
L
a
n
g
u
a
g
e
s
,
2
0
1
0
,
p
p
.
5
2
-
5
6
.
[2
0
]
M
.
Ou
tah
a
jala
,
L.
Zek
o
u
a
r,
a
n
d
P
.
Ro
ss
o
,
“
Bu
il
d
i
n
g
a
n
a
n
n
o
tate
d
c
o
rp
u
s
fo
r
Am
a
z
ig
h
e
”
.
In
W
il
l
a
p
p
e
a
r
In
Pro
c
.
o
f
4
th
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Ama
zig
h
a
n
d
IC
T
,
2
0
1
1
.
[2
1
]
R.
M
.
Ha
n
ifa
e
t
a
l.
,
“
Vo
ice
d
a
n
d
u
n
v
o
ice
d
se
p
a
ra
ti
o
n
i
n
m
a
lay
sp
e
e
c
h
u
sin
g
z
e
ro
c
r
o
ss
in
g
ra
te
a
n
d
e
n
e
rg
y
,”
In
d
o
n
e
sia
n
J
o
u
r
n
a
l
o
f
El
e
c
tric
a
l
En
g
i
n
e
e
rin
g
a
n
d
C
o
mp
u
ter
S
c
ien
c
e
,
v
o
l
.
1
6
,
n
o
.
2
,
p
p
.
775
-
7
8
0
,
2
0
1
9
,
d
o
i:
1
0
.
1
1
5
9
1
/
ij
e
e
c
s.v
1
6
.
i2
.
p
p
7
7
5
-
7
8
0
.
[2
2
]
S
.
Ho
c
h
re
it
e
r
a
n
d
J.
S
c
h
m
i
d
h
u
b
e
r
,
“
Lo
n
g
S
h
o
rt
-
Term
M
e
m
o
ry
,”
Ne
u
ra
l
Co
mp
u
t
a
ti
o
n
,
v
o
l
.
9
,
n
o
.
8
,
p
p
.
1
7
3
5
-
1
7
8
0
,
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
.
[2
3
]
M
.
Zu
l
q
a
rn
a
in
,
R.
G
h
a
z
a
li
,
Y.
M
.
M
o
h
m
a
d
Ha
ss
im,
a
n
d
M
.
R
e
h
a
n
,
“
A
c
o
m
p
a
ra
ti
v
e
re
v
iew
o
n
d
e
e
p
lea
rn
i
n
g
m
o
d
e
ls
fo
r
tex
t
c
las
sifica
ti
o
n
,”
In
d
o
n
e
si
a
n
J
o
u
rn
a
l
o
f
El
e
c
trica
l
E
n
g
i
n
e
e
rin
g
a
n
d
C
o
mp
u
ter
S
c
ien
c
e
,
v
o
l
.
1
9
,
n
o
.
1
,
p
p
.
3
2
5
-
3
3
5
,
2
0
2
0
,
d
o
i:
1
0
.
1
1
5
9
1
/i
jee
c
s.v
1
9
.
i1
.
p
p
3
2
5
-
3
3
5
.
[2
4
]
I.
Ja
m
a
led
d
y
n
a
n
d
M
.
Bin
iz,
“
C
o
n
tri
b
u
t
io
n
t
o
Ara
b
ic
Te
x
t
Clas
s
ifi
c
a
ti
o
n
Us
in
g
M
a
c
h
in
e
Lea
rn
in
g
Tec
h
n
iq
u
e
s
,”
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
B
u
si
n
e
ss
In
telli
g
e
n
c
e
,
p
p
.
1
8
-
32
,
2
0
2
1
,
d
o
i:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
7
6
5
0
8
-
8
_
2
.
[2
5
]
C.
Be
rg
m
e
ir,
R.
J.
H
y
n
d
m
a
n
,
a
n
d
B.
Ko
o
,
“
A
n
o
te
o
n
th
e
v
a
li
d
it
y
o
f
c
ro
ss
-
v
a
li
d
a
ti
o
n
fo
r
e
v
a
lu
a
ti
n
g
a
u
to
re
g
re
ss
iv
e
ti
m
e
se
ries
p
re
d
icti
o
n
,
”
Co
mp
u
ta
ti
o
n
a
l
S
t
a
ti
stics
&
Da
t
a
An
a
lys
is
,
v
o
l.
1
2
0
,
p
p
.
7
0
-
8
3
,
A
p
r.
2
0
1
8
,
d
o
i:
1
0
.
1
0
1
6
/j
.
c
sd
a
.
2
0
1
7
.
1
1
.
0
0
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
24
,
No
.
3
,
Dec
em
b
er
2
0
2
1
:
1
8
1
4
-
1
8
2
2
1822
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
O
tm
a
n
M
a
a
r
o
u
f
:
re
c
e
iv
e
d
h
is
m
a
ste
r'
s
d
e
g
re
e
in
b
u
si
n
e
ss
in
te
ll
ig
e
n
c
e
in
2
0
1
8
fro
m
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
o
l
o
g
y
,
Un
i
v
e
rsity
S
u
lt
a
n
M
o
u
lay
S
li
m
a
n
Be
n
i
-
M
e
ll
a
l.
He
i
s
c
u
rre
n
tl
y
a
P
h
D
d
e
g
re
e
su
t
d
e
n
t.
His
re
se
a
rc
h
a
c
ti
v
it
ies
a
re
lo
c
a
ted
in
th
e
a
re
a
o
f
n
a
tu
ra
l
lan
g
u
a
g
e
p
r
o
c
e
ss
in
g
s
p
e
c
ifi
c
a
ll
y
;
it
d
e
a
ls
with
P
a
rt
o
f
sp
e
e
c
h
,
n
a
m
e
d
e
n
ti
ty
re
c
o
g
n
it
i
o
n
,
a
n
d
li
m
a
ti
z
a
ti
o
n
-
ste
a
m
in
g
o
f
t
h
e
Am
a
z
ig
h
lan
g
u
a
g
e
.
Ra
c
h
id
El
Ay
a
c
h
i
o
b
tain
e
d
a
d
e
g
re
e
in
M
a
ste
r
o
f
I
n
fo
rm
a
ti
c
Tele
c
o
m
a
n
d
M
u
lt
ime
d
ia
(IT
M
)
in
2
0
0
6
fro
m
th
e
F
a
c
u
lt
y
of
S
c
ien
c
e
s,
M
o
h
a
m
m
e
d
V
Un
i
v
e
rsity
(M
o
ro
c
c
o
)
a
n
d
a
P
h
.
D.
d
e
g
re
e
in
c
o
m
p
u
ter
sc
ien
c
e
fro
m
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
o
lo
g
y
,
S
u
lt
a
n
M
o
u
la
y
S
li
m
a
n
e
Un
iv
e
rsit
y
(
M
o
r
o
c
c
o
).
He
is
c
u
rre
n
tl
y
a
m
e
m
b
e
r
o
f
lab
o
ra
t
o
ry
T
IAD
a
n
d
a
p
ro
fe
ss
o
r
a
t
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
o
lo
g
y
,
S
u
lt
a
n
M
o
u
lay
S
li
m
a
n
e
Un
iv
e
rsit
y
,
M
o
r
o
c
c
o
.
His
re
se
a
rc
h
fo
c
u
se
s
o
n
ima
g
e
p
r
o
c
e
ss
in
g
,
p
a
tt
e
r
n
re
c
o
g
n
it
i
o
n
,
m
a
c
h
in
e
lea
rn
i
n
g
,
a
n
d
se
m
a
n
ti
c
we
b
.
Mo
h
a
m
e
d
Bin
iz
:
re
c
e
iv
e
d
h
is
m
a
ste
r'
s
d
e
g
re
e
in
b
u
sin
e
ss
in
te
ll
ig
e
n
c
e
i
n
2
0
1
4
a
n
d
P
h
.
D
d
e
g
re
e
in
c
o
m
p
u
ter
sc
ien
c
e
i
n
2
0
1
8
fr
o
m
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
o
l
o
g
y
,
U
n
iv
e
rsit
y
S
u
lt
a
n
M
o
u
lay
S
li
m
a
n
Be
n
i
-
M
e
ll
a
l.
He
is
a
p
ro
fe
ss
o
r
a
t
p
o
l
y
d
isc
ip
li
n
a
ry
fa
c
u
lt
y
U
n
iv
e
rsit
y
S
u
lt
a
n
M
y
S
li
m
a
n
e
Be
n
i
M
e
ll
a
l
m
o
ro
c
c
o
.
His
re
se
a
rc
h
a
c
ti
v
it
ies
a
re
lo
c
a
ted
in
th
e
a
re
a
o
f
th
e
se
m
a
n
ti
c
we
b
e
n
g
in
e
e
rin
g
a
n
d
d
e
e
p
lea
rn
in
g
sp
e
c
ifi
c
a
ll
y
,
it
d
e
a
ls
with
th
e
re
se
a
rc
h
q
u
e
sti
o
n
o
f
th
e
e
v
o
l
u
ti
o
n
o
f
o
n
t
o
l
o
g
y
,
b
ig
d
a
t
a
,
n
a
tu
ra
l
la
n
g
u
a
g
e
p
r
o
c
e
ss
in
g
,
a
n
d
d
y
n
a
m
ic p
r
o
g
ra
m
m
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.