I
nte
rna
t
io
na
l J
o
urna
l o
f
Adv
a
nces in Applie
d
Science
s
(
I
J
AAS)
Vo
l.
1
5
,
No
.
2
,
J
u
n
e
20
2
6
,
p
p
.
687
~
6
9
5
I
SS
N:
2252
-
8
8
1
4
,
DOI
:
1
0
.
1
1
5
9
1
/ijaas
.
v
1
5
.
i
2
.
pp
687
-
6
9
5
687
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
a
s
.
ia
esco
r
e.
co
m
Ass
ess
ment of
dee
p learning
bas
ed
H
indi
O
dia
bidire
ctiona
l
ma
chine t
ra
nsla
ti
o
n sy
stem
Su
bh
a
s
hree
Sa
t
pa
t
hy
1
,
Sm
it
a
pra
v
a
M
is
hra
2
,
Aj
it
K
um
a
r
Na
y
a
k
2
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
S
i
k
s
h
a
‘
O
’
A
n
u
s
a
n
d
h
a
n
(
D
e
e
me
d
t
o
b
e
U
n
i
v
e
r
si
t
y
)
,
B
h
u
b
a
n
e
sw
a
r
,
I
n
d
i
a
2
D
e
p
a
r
t
m
e
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
S
i
k
s
h
a
‘
O
’
A
n
u
s
a
n
d
h
a
n
(
D
e
e
m
e
d
t
o
b
e
U
n
i
v
e
r
s
i
t
y
)
,
B
h
u
b
a
n
e
sw
a
r
,
I
n
d
i
a
Art
icle
I
n
fo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Oct
3
1
,
2
0
2
5
R
ev
is
ed
Ma
y
1
,
2
0
2
6
Acc
ep
ted
Ma
y
2
2
,
2
0
2
6
In
d
ia
is
a
v
a
st
n
a
ti
o
n
with
a
d
i
v
e
rse
ra
n
g
e
o
f
c
u
lt
u
re
s
a
n
d
lan
g
u
a
g
e
s.
M
o
st
In
d
ian
s
c
h
o
o
se
t
o
u
se
th
e
ir
n
a
t
iv
e
lan
g
u
a
g
e
s
w
h
e
n
c
o
m
m
u
n
ica
ti
n
g
with
m
a
c
h
in
e
s.
To
i
n
teg
ra
te
sm
a
rt
tec
h
n
o
l
o
g
ies
i
n
to
e
v
e
ry
fa
c
e
t
o
f
In
d
ia
n
so
c
iety
,
e
fficie
n
t
sy
ste
m
s
th
a
t
c
a
n
p
o
sit
i
v
e
ly
i
d
e
n
ti
fy
I
n
d
ia
n
lan
g
u
a
g
e
s
m
u
st
b
e
e
sta
b
li
sh
e
d
.
M
a
c
h
in
e
t
ra
n
sla
ti
o
n
(M
T)
stu
d
ies
c
o
m
p
rise
m
o
st
o
f
t
h
e
n
a
tu
ra
l
lan
g
u
a
g
e
p
ro
c
e
ss
in
g
(NLP
)
in
th
e
e
ra
o
f
m
u
lt
i
li
n
g
u
a
l
c
o
m
p
u
t
e
r
-
h
u
m
a
n
in
tera
c
ti
o
n
.
Ti
ll
n
o
w
,
les
s
e
m
p
h
a
sis
h
a
s
b
e
e
n
p
lac
e
d
t
o
d
e
v
e
lo
p
M
T
sy
ste
m
s
a
m
o
n
g
I
n
d
ian
l
a
n
g
u
a
g
e
s.
Ye
t
a
g
a
in
,
b
u
il
d
in
g
a
q
u
a
li
tativ
e
a
n
d
q
u
a
n
ti
tati
v
e
c
o
rp
u
s
i
n
th
e
se
lan
g
u
a
g
e
s
is
c
h
a
ll
e
n
g
i
n
g
.
T
h
is
wo
r
k
fo
c
u
se
s
o
n
two
In
d
ic
lan
g
u
a
g
e
s
fo
r
t
h
e
d
e
v
e
lo
p
m
e
n
t
o
f
a
Hin
d
i
to
O
d
ia
b
id
irec
ti
o
n
a
l
m
a
c
h
in
e
tran
sla
ti
o
n
sy
ste
m
(HO
BM
T).
Bil
in
g
u
a
l
e
v
a
l
u
a
ti
o
n
u
n
d
e
rstu
d
y
(BLE
U),
wo
rd
e
rro
r
ra
te
(W
ER),
c
h
a
ra
c
ter
e
rro
r
ra
te
(CER),
a
n
d
m
e
tri
c
fo
r
e
v
a
lu
a
ti
o
n
o
f
tran
sla
ti
o
n
wi
th
e
x
p
li
c
it
o
r
d
e
rin
g
(
M
ET
EOR)
e
v
a
l
u
a
ti
o
n
m
e
tri
c
s
a
re
u
se
d
to
a
ss
e
ss
th
e
a
c
c
u
ra
c
y
o
f
th
e
tra
n
sla
ti
o
n
.
Th
e
m
o
st
a
d
v
a
n
c
e
d
se
q
u
e
n
ti
a
l
d
e
e
p
lea
rn
in
g
(DL)
m
o
d
e
ls,
su
c
h
a
s
re
c
u
rre
n
t
n
e
u
ra
l
n
e
tw
o
rk
(RNN
),
l
o
n
g
sh
o
rt
-
term
m
e
m
o
ry
(LS
T
M
),
a
n
d
g
a
t
e
d
re
c
u
rre
n
t
u
n
it
(G
RU),
a
re
u
s
e
d
in
th
is
stu
d
y
.
I
n
th
is
re
se
a
rc
h
,
RNN
is
o
b
se
rv
e
d
wit
h
imp
r
o
v
e
d
tra
n
sla
ti
o
n
re
su
lt
s
d
u
e
t
o
it
s se
q
u
e
n
ti
a
l
d
a
ta h
a
n
d
li
n
g
with
c
o
n
tex
t
p
re
se
rv
a
ti
o
n
.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
E
v
alu
atio
n
m
etr
ics
I
n
d
ic
lan
g
u
ag
es
Ma
ch
in
e
tr
an
s
latio
n
Natu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
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
:
Su
b
h
ash
r
ee
Satp
ath
y
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
Sik
s
h
a
‘
O’
An
u
s
an
d
h
a
n
(
Dee
m
ed
to
b
e
Un
iv
er
s
ity
)
B
h
u
b
an
eswar
,
Od
is
h
a
7
5
1
0
3
0
,
I
n
d
ia
E
m
ail: su
b
h
ash
r
ee
s
atp
ath
y
1
6
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Ma
ch
in
e
tr
an
s
latio
n
(
MT
)
is
k
n
o
wn
as
a
p
o
r
tio
n
o
f
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
(
N
L
P)
wh
ich
au
to
m
atica
lly
co
n
v
er
ts
tex
t
o
r
s
p
ee
ch
b
etwe
en
d
if
f
er
en
t
la
n
g
u
ag
es
u
s
in
g
alg
o
r
ith
m
s
,
s
t
atis
tical
m
o
d
els,
o
r
n
eu
r
al
n
etwo
r
k
s
to
ac
h
iev
e
a
cc
u
r
ate
an
d
co
n
tex
t
-
awa
r
e
tr
an
s
latio
n
s
.
I
n
d
ia
h
as
a
wid
e
v
ar
iety
o
f
s
p
o
k
e
n
lan
g
u
ag
es,
wh
er
e
m
o
r
e
th
an
a
th
o
u
s
an
d
d
ialec
ts
an
d
at
least
3
0
d
is
tin
ct
lan
g
u
ag
es
c
o
ex
is
t.
MT
s
y
s
tem
s
ar
e
v
ital
f
o
r
o
v
er
co
m
i
n
g
lan
g
u
ag
e
b
ar
r
ier
s
an
d
p
r
o
v
id
in
g
d
ev
el
o
p
m
en
ts
lik
e
Dig
ital
I
n
d
ia.
Seq
u
en
ce
-
to
-
s
eq
u
en
ce
(
s
eq
2
s
eq
)
m
o
d
els
h
av
e
co
n
s
id
er
ab
ly
en
r
ich
ed
th
e
tr
an
s
latio
n
q
u
ality
b
y
ca
p
tu
r
in
g
lan
g
u
ag
e
d
ata
co
n
tex
tu
a
l
d
ep
en
d
e
n
cies
in
d
ee
p
lear
n
in
g
(
DL
)
tech
n
iq
u
es
[
1
]
,
[
2
]
.
I
n
d
ia
n
l
a
n
g
u
ag
es
t
h
a
t
l
ac
k
co
m
p
u
t
ati
o
n
al
m
o
d
e
ls
,
an
n
o
ta
te
d
d
at
ase
ts
,
li
n
g
u
is
tic
to
o
l
s
,
a
n
d
o
t
h
e
r
d
i
g
it
al
r
es
o
u
r
ce
s
f
o
r
tas
k
s
l
ik
e
N
L
P
ar
e
r
ef
e
r
r
ed
t
o
as
lo
w
-
r
es
o
u
r
ce
I
n
d
ic
la
n
g
u
a
g
es
[
3
]
.
W
h
en
c
o
m
p
ar
e
d
t
o
in
te
r
n
at
io
n
al
la
n
g
u
ag
es
wit
h
ab
u
n
d
a
n
t
r
eso
u
r
c
es
l
ik
e
E
n
g
l
is
h
o
r
F
r
e
n
ch
,
th
ese
lo
w
-
r
eso
u
r
ce
la
n
g
u
a
g
es
f
r
e
q
u
e
n
t
ly
l
ac
k
s
u
f
f
i
cie
n
t
c
o
r
p
o
r
a
,
d
ic
ti
o
n
ar
i
es,
a
n
d
r
ese
a
r
c
h
atte
n
t
io
n
.
E
x
a
m
p
les
i
n
cl
u
d
i
n
g
Ma
n
ip
u
r
i
,
Ko
n
k
a
n
i
,
S
a
n
ta
li,
B
o
d
o
,
a
n
d
O
d
i
a
[
4
]
,
[
5
]
.
Dev
elo
p
in
g
r
eso
u
r
ce
s
f
o
r
th
e
s
e
lan
g
u
ag
es
is
v
ital
f
o
r
p
r
e
s
er
v
in
g
lin
g
u
is
tic
d
iv
er
s
ity
a
n
d
aid
i
n
g
ac
ce
s
s
to
tech
n
o
lo
g
y
f
o
r
s
p
ea
k
er
s
o
f
s
u
ch
lan
g
u
ag
e
s
.
Hin
d
i
is
an
I
n
d
o
-
A
r
y
an
lan
g
u
a
g
e
s
p
o
k
en
p
r
im
a
r
ily
in
I
n
d
ia
a
n
d
g
lo
b
ally
,
with
o
v
er
6
0
0
m
illi
o
n
s
p
ea
k
e
r
s
,
in
clu
d
i
n
g
n
ativ
e
a
n
d
s
ec
o
n
d
-
lan
g
u
a
g
e
s
p
ea
k
er
s
.
I
n
d
ia'
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
2
,
J
u
n
e
2
0
2
6
:
6
8
7
-
695
688
liter
atu
r
e,
cu
ltu
r
e,
m
ed
ia,
a
n
d
ed
u
ca
tio
n
all
d
ep
en
d
o
n
it
[
6
]
.
I
t
is
wr
itten
in
th
e
Dev
an
ag
ar
i
s
cr
ip
t
an
d
h
as
it
s
r
o
o
ts
in
San
s
k
r
it,
with
s
ig
n
if
i
ca
n
t
in
f
lu
en
ce
s
f
r
o
m
Per
s
ian
,
Ar
ab
ic,
an
d
E
n
g
lis
h
.
Hin
d
i
is
o
n
e
o
f
th
e
o
f
f
icial
lan
g
u
ag
es
o
f
th
e
g
o
v
er
n
m
en
t o
f
I
n
d
ia.
T
h
e
I
n
d
o
-
Ar
y
an
lan
g
u
ag
e
k
n
o
w
n
as
Od
ia
is
m
ain
ly
s
p
o
k
en
in
th
e
s
tate
o
f
Od
is
h
a
an
d
its
n
eig
h
b
o
r
in
g
r
eg
io
n
s
.
I
t
is
o
n
e
o
f
I
n
d
ia'
s
2
2
s
ch
ed
u
led
lan
g
u
a
g
es
an
d
was
d
esig
n
ated
a
6
th
class
ical
lan
g
u
ag
e
o
f
I
n
d
ia
d
u
e
to
its
r
ich
liter
ar
y
h
er
itag
e
a
n
d
h
is
to
r
ical
s
ig
n
if
ican
ce
.
I
t
h
as
its
o
wn
s
cr
ip
t
an
d
h
as
a
h
is
to
r
y
o
v
er
m
o
r
e
th
an
1
,
0
0
0
y
ea
r
s
.
Od
ia
is
r
en
o
wn
ed
f
o
r
its
co
n
tr
ib
u
tio
n
s
to
I
n
d
ian
cu
ltu
r
e,
d
ev
o
tio
n
al
p
o
etr
y
,
an
d
an
cien
t
tex
ts
.
T
h
is
lan
g
u
a
g
e,
s
p
o
k
en
b
y
m
o
r
e
th
an
4
5
m
illi
o
n
p
eo
p
le,
is
a
n
im
p
o
r
tan
t
p
ar
t
o
f
t
h
e
r
eg
io
n
'
s
id
en
tity
,
ed
u
ca
t
io
n
,
an
d
g
o
v
er
n
m
e
n
t
[
7
]
.
Ma
n
y
Od
is
h
a
citizen
s
s
till
h
av
e
tr
o
u
b
le
g
r
asp
i
n
g
in
f
o
r
m
atio
n
wr
itten
in
Hin
d
i,
d
esp
ite
Hin
d
i
b
ein
g
th
e
o
f
f
ici
al
lan
g
u
a
g
e
o
f
I
n
d
ia.
I
n
th
is
c
ase,
a
s
u
itab
le
MT
s
y
s
tem
f
r
o
m
Hin
d
i to
O
d
ia
will e
n
ab
le
p
eo
p
le
to
co
m
p
r
e
h
en
d
an
d
u
tili
ze
Hin
d
i m
o
r
e
cr
ea
tiv
e
way
[
8
]
,
[
9
]
.
R
u
le
-
b
ased
MT
,
s
tatis
tical
M
T
,
n
eu
r
al
MT
,
e
x
am
p
le
-
b
ased
MT,
an
d
hy
b
r
id
MT
a
r
e
th
e
ca
teg
o
r
ies
o
f
MT
s
y
s
tem
s
[
2
]
,
[
4
]
.
Neu
r
al
MT
h
as
d
em
o
n
s
tr
ated
p
r
o
m
is
e
in
h
an
d
lin
g
s
eq
u
en
tial
d
ata
an
d
en
h
an
ci
n
g
tr
an
s
latio
n
ac
cu
r
ac
y
,
esp
ec
ially
with
r
ec
u
r
r
en
t
ar
ch
itectu
r
es
lik
e
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
(
R
NN)
[
1
0
]
,
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
[
1
1
]
,
a
n
d
g
ate
d
r
ec
u
r
r
en
t
u
n
it
(
GR
U)
[
1
2
]
.
Alth
o
u
g
h
p
ar
alle
l
co
r
p
o
r
a
f
o
r
I
n
d
ic
lan
g
u
ag
es
h
av
e
b
ee
n
ex
p
a
n
d
ed
b
y
r
ec
en
t
in
itiativ
es
lik
e
th
e
Sam
an
an
tar
d
ataset,
n
o
is
e
an
d
in
co
n
s
is
ten
cy
p
r
o
b
lem
s
s
till
ex
is
t
[
1
3
]
.
DL
m
o
d
els
f
o
r
Hin
d
i
-
Od
ia
tr
an
s
latio
n
h
av
e
n
o
t
y
et
b
ee
n
th
o
r
o
u
g
h
ly
ev
al
u
ated
,
th
o
u
g
h
MT
f
o
r
Hin
d
i
-
E
n
g
lis
h
o
r
o
t
h
er
I
n
d
ic
p
air
s
h
as
b
ee
n
t
h
e
s
u
b
ject
o
f
n
u
m
er
o
u
s
s
tu
d
ie
s
.
I
n
c
o
u
n
tr
ies
with
m
u
ltip
le
lan
g
u
a
g
es,
lik
e
I
n
d
i
a,
it
is
r
eq
u
ir
ed
to
h
av
e
a
l
o
t
o
f
h
i
g
h
-
q
u
ality
p
ar
allel
d
atasets
f
o
r
I
n
d
ia
n
lan
g
u
ag
es
t
o
b
u
ild
lo
w
-
r
eso
u
r
ce
I
n
d
ic
MT
s
y
s
tem
s
[
1
4
]
.
R
esear
ch
o
n
b
id
ir
ec
tio
n
al
Hin
d
i
-
Od
ia
MT
s
y
s
tem
s
is
s
im
p
ly
lack
in
g
.
So
,
co
m
p
ar
ati
v
e
an
aly
s
is
o
f
DL
m
o
d
els
o
n
s
u
ch
d
atasets
i
s
r
ar
e.
I
t
is
g
o
i
n
g
to
b
e
d
if
f
icu
lt
to
id
en
tify
th
e
b
est
ar
ch
itectu
r
es
f
o
r
lo
w
-
r
eso
u
r
ce
I
n
d
ic
la
n
g
u
ag
es
with
o
u
t
tr
u
s
tf
u
l
ev
alu
atio
n
.
Ho
wev
er
,
r
esear
ch
er
s
h
av
e
d
ev
elo
p
e
d
th
e
Sam
an
an
tar
d
ataset,
a
v
ast p
ar
allel
d
ataset
co
llectio
n
f
o
r
I
n
d
ian
l
an
g
u
ag
es.
I
n
th
is
wo
r
k
,
Hin
d
i
an
d
Od
ia
lan
g
u
ag
e
d
ata
s
et
s
ar
e
co
n
s
id
er
ed
to
b
u
ild
a
Hin
d
i
to
Od
ia
b
i
d
ir
ec
tio
n
al
m
ac
h
in
e
tr
a
n
s
latio
n
s
y
s
tem
(
HOBMT
)
.
T
h
e
f
ir
s
t
s
tep
o
f
th
is
wo
r
k
t
o
cr
ea
te
a
Hin
d
i
-
Od
i
a
p
ar
allel
c
o
r
p
u
s
o
f
o
v
er
6
0
,
0
0
0
s
en
ten
ce
p
air
s
.
I
t
in
v
esti
g
ates h
o
w
d
if
f
er
e
n
t D
L
m
eth
o
d
s
lik
e
R
NN,
L
STM
,
an
d
GR
U
p
er
f
o
r
m
o
n
Hin
d
i
-
Od
ia
b
id
ir
ec
tio
n
al
tr
an
s
latio
n
.
All
th
e
tr
an
s
lat
io
n
s
a
r
e
ev
alu
ated
u
s
in
g
s
tan
d
ar
d
ev
alu
atio
n
m
etr
ics
,
i.e
.
,
b
ilin
g
u
al
ev
alu
atio
n
u
n
d
e
r
s
tu
d
y
(
B
L
E
U)
[
1
5
]
,
m
etr
ic
f
o
r
ev
alu
atio
n
o
f
tr
an
s
latio
n
w
ith
ex
p
licit
o
r
d
er
in
g
(
ME
T
E
OR
)
[
1
6
]
,
wo
r
d
er
r
o
r
r
ate
(
W
E
R
)
[
1
7
]
,
an
d
ch
a
r
ac
ter
er
r
o
r
r
ate
(
C
E
R
)
[
1
8
]
.
T
r
a
n
s
latio
n
q
u
ality
i
s
ev
alu
ated
with
v
ar
y
in
g
p
ar
a
m
eter
s
ettin
g
s
,
lik
e
lear
n
in
g
r
ate,
b
atch
s
ize,
an
d
ep
o
c
h
s
,
o
n
th
e
tr
an
s
latio
n
p
er
f
o
r
m
an
ce
.
I
t
ca
teg
o
r
izes
t
h
e
b
en
e
f
its
an
d
ch
allen
g
es
o
f
ea
ch
m
o
d
el
in
p
r
o
ce
s
s
in
g
s
im
p
le
an
d
s
h
o
r
t
s
en
ten
ce
s
o
f
a
lo
w
-
r
eso
u
r
ce
d
at
aset.
T
h
is
wo
r
k
f
o
cu
s
es o
n
b
u
ild
in
g
Hin
d
i
-
Od
ia
MT
s
y
s
tem
alo
n
g
with
d
ataset
cr
ea
tio
n
.
I
t
e
m
p
h
asizes
th
e
im
p
o
r
tan
ce
o
f
a
p
p
lied
s
cien
ce
s
a
n
d
in
f
o
r
m
atio
n
tech
n
o
lo
g
ies
b
y
tack
lin
g
is
s
u
es
in
lo
w
r
eso
u
r
ce
lan
g
u
ag
es
an
d
a
d
v
an
cin
g
cr
o
s
s
-
lin
g
u
al
co
m
m
u
n
icatio
n
tech
n
o
lo
g
ies
in
th
e
d
ig
ital
ec
o
s
y
s
tem
.
Mo
r
eo
v
er
,
it
p
r
o
m
o
tes
d
ig
ital
in
clu
s
io
n
b
y
p
r
o
v
i
d
in
g
s
p
ea
k
e
r
s
o
f
u
n
d
er
r
ep
r
esen
ted
lan
g
u
a
g
es
with
im
p
r
o
v
ed
ac
ce
s
s
to
g
o
v
er
n
m
en
t
s
er
v
i
ce
s
,
d
ig
ital
co
n
ten
t,
en
g
in
e
er
in
g
,
in
f
o
r
m
atio
n
tech
n
o
lo
g
y
,
an
d
d
i
f
f
er
en
t
ed
u
ca
tio
n
al
m
ater
ials
b
y
co
n
c
en
tr
atin
g
o
n
Hin
d
i
-
Od
ia
MT
s
y
s
tem
.
T
h
e
r
est
o
f
th
e
ar
ticle
is
el
ab
o
r
ated
as
f
o
llo
ws.
Sectio
n
2
is
th
e
p
r
im
ar
y
p
ar
t
o
f
th
e
m
eth
o
d
,
co
m
p
r
is
ed
with
th
e
d
ataset
co
llectio
n
an
d
p
r
e
p
ar
atio
n
,
d
if
f
er
en
t
tech
n
i
q
u
es
,
in
clu
d
i
n
g
h
y
p
e
r
p
ar
am
ete
r
s
ettin
g
s
wit
h
f
lo
w
o
f
wo
r
k
.
Sectio
n
3
u
n
f
o
ld
s
th
e
e
x
p
er
im
e
n
tal
r
esu
lt
s
an
d
d
is
cu
s
s
io
n
s
.
S
ec
tio
n
4
co
n
clu
d
es
th
e
wo
r
k
a
n
d
h
i
g
h
lig
h
ts
its
f
u
t
u
r
e
p
er
s
p
ec
tiv
es.
2.
M
E
T
H
O
D
2
.
1
.
Da
t
a
s
et
prepa
ra
t
io
n
I
t
is
ea
s
y
to
d
e
v
elo
p
a
b
ilin
g
u
al
co
r
p
u
s
f
o
r
r
eso
u
r
ce
-
r
ic
h
lan
g
u
a
g
e
p
air
s
,
lik
e
E
n
g
lis
h
,
C
h
in
ese,
J
ap
an
ese,
Ger
m
an
,
an
d
s
o
f
o
r
th
,
b
ec
au
s
e
th
e
r
e
a
r
e
p
le
n
t
y
o
f
r
eso
u
r
ce
s
av
ailab
le
f
o
r
co
r
p
o
r
a
co
llectio
n
.
T
h
er
e
ar
e
s
o
m
an
y
web
s
ites
,
b
o
o
k
s
,
g
o
v
er
n
m
en
t,
an
d
in
s
tit
u
tio
n
al
d
atasets
av
ailab
le
f
o
r
d
ata
co
llectio
n
[
1
9
]
.
Fo
r
d
ata
p
r
ep
ar
atio
n
,
it is
im
p
o
r
tan
t to
h
av
e
an
ex
ten
s
iv
e
d
at
aset.
Pre
p
r
o
ce
s
s
in
g
an
d
clea
n
in
g
o
f
th
ese
d
ata
ar
e
ess
en
tial to
m
ain
tain
th
at
th
e
m
o
d
els ca
n
tr
ain
ef
f
ec
tiv
el
y
.
T
h
e
r
aw
d
ata
f
r
o
m
th
e
la
n
g
u
ag
e
in
clu
d
e
s
n
o
t o
n
l
y
a
lo
t
o
f
d
ata
f
r
o
m
o
th
er
lan
g
u
a
g
es
b
u
t
also
o
th
er
r
an
d
o
m
ch
ar
ac
ter
s
.
T
h
e
len
g
th
o
f
th
e
s
e
n
ten
ce
s
also
v
ar
ies
s
ig
n
if
ican
tly
.
T
h
u
s
,
it
n
ee
d
s
to
b
e
s
tan
d
a
r
d
ized
to
t
r
ain
a
s
o
u
n
d
m
o
d
el.
Do
wn
lo
ad
i
n
g
c
o
n
te
n
ts
f
r
o
m
web
s
ites
,
alig
n
in
g
o
f
d
o
c
u
m
en
t
s
an
d
s
e
n
ten
ce
s
,
an
d
p
o
s
t
-
p
r
o
ce
s
s
in
g
ar
e
th
e
s
tep
s
n
ee
d
to
b
e
f
o
llo
wed
to
p
r
ep
a
r
e
th
e
d
ataset.
B
esid
es
th
is
,
d
ep
en
d
i
n
g
o
n
th
e
tr
aits
o
f
ea
ch
la
n
g
u
ag
e
p
air
,
a
f
ew
ap
p
r
o
p
r
iate
p
r
o
ce
s
s
in
g
s
tep
s
ca
n
b
e
ad
d
e
d
[
2
0
]
,
[
2
1
]
.
Dev
el
o
p
i
n
g
a
n
MT
s
y
s
te
m
i
n
I
n
d
ia
n
l
a
n
g
u
a
g
e
is
a
v
e
r
y
to
u
g
h
tas
k
.
B
ec
a
u
s
e
it h
as
m
u
l
ti
p
le
i
s
s
u
es li
k
e
s
cr
i
p
t
c
o
m
p
le
x
it
y
,
m
o
r
p
h
o
l
o
g
ical
r
ic
h
n
ess
,
l
ac
k
o
f
s
t
an
d
ar
d
iz
ati
o
n
,
s
p
ee
c
h
r
e
co
g
n
it
io
n
i
s
s
u
es
,
d
a
ta
s
ca
r
c
it
y
,
an
d
l
im
i
te
d
r
es
ea
r
c
h
s
u
p
p
o
r
t
.
E
v
e
n
th
o
u
g
h
th
e
r
e
ar
e
n
u
m
e
r
o
u
s
m
a
jo
r
r
e
ce
n
t
ad
v
an
c
em
e
n
ts
i
n
c
o
r
p
u
s
d
e
v
e
lo
p
m
e
n
t
,
m
o
r
e
ef
f
o
r
ts
s
h
o
u
l
d
b
e
m
a
d
e
t
o
c
r
ea
te
la
r
g
e
-
s
c
a
le
d
i
v
e
r
s
i
f
i
ed
d
at
ase
t
f
o
r
a
ll
I
n
d
ic
l
an
g
u
a
g
es
[
1
3
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
A
s
s
es
s
men
t o
f d
ee
p
lea
r
n
in
g
b
a
s
ed
Hin
d
i O
d
ia
b
id
ir
ec
tio
n
a
l
ma
ch
in
e
tr
a
n
s
la
tio
n
…
(
S
u
b
h
a
s
h
r
ee
S
a
tp
a
th
y)
689
I
n
d
ian
in
s
titu
tio
n
s
,
r
esear
ch
e
r
s
,
an
d
o
r
g
a
n
izatio
n
s
h
av
e
b
e
g
u
n
d
e
v
elo
p
in
g
MT
s
y
s
tem
s
f
o
r
I
n
d
ian
lan
g
u
ag
es
an
d
h
a
v
e
ac
h
iev
ed
s
atis
f
ac
to
r
y
r
esu
lts
[
2
2
]
.
B
u
t
Hin
d
i
-
Od
ia
is
a
lo
w
-
r
eso
u
r
ce
lan
g
u
ag
e
p
air
,
an
d
d
ev
elo
p
in
g
a
b
ilin
g
u
al
co
r
p
u
s
is
d
if
f
icu
lt.
V
er
y
less
b
ilin
g
u
al
elec
tr
o
n
ic
d
o
c
u
m
en
ts
ar
e
p
r
esen
t
in
o
n
lin
e
p
latf
o
r
m
.
T
h
e
w
o
r
k
is
to
b
u
ild
a
Hin
d
i
to
Od
ia
b
ilin
g
u
al
MT
s
y
s
tem
u
s
in
g
d
if
f
er
en
t
DL
m
eth
o
d
s
.
T
h
e
au
th
o
r
s
co
llected
m
o
r
e
th
a
n
o
n
e
lak
h
Hin
d
i
to
Od
ia
s
en
ten
ce
p
air
s
.
T
h
e
co
llectio
n
p
r
o
ce
s
s
is
q
u
i
te
ted
io
u
s
,
an
d
th
e
v
er
if
icatio
n
p
r
o
ce
s
s
m
ak
es
it
ev
en
m
o
r
e
d
if
f
ic
u
lt.
Hu
m
an
v
alid
atio
n
is
n
ee
d
ed
f
o
r
ea
c
h
s
o
u
r
ce
to
tar
g
et
lan
g
u
ag
e
alig
n
m
e
n
t.
Pro
p
o
s
e
d
d
atasets
ar
e
co
llected
f
r
o
m
th
e
web
s
ite
“h
ttp
s
://www.
m
an
y
th
in
g
s
.
o
r
g
/an
k
i/”.
T
h
e
d
atasets
ar
e
in
in
ter
n
atio
n
al
lan
g
u
ag
e
f
o
r
m
[
2
3
]
.
T
h
e
E
n
g
lis
h
s
en
ten
ce
s
h
av
e
b
ee
n
tr
an
s
lated
m
an
u
ally
in
to
two
d
if
f
e
r
en
t I
n
d
ic
lan
g
u
a
g
es,
Hin
d
i a
n
d
O
d
ia.
T
ab
le
1
r
ep
r
esen
ts
o
n
e
ex
am
p
le
o
f
p
r
o
p
o
s
ed
d
ataset.
T
ab
le
1
.
E
x
am
p
le
o
f
p
r
o
p
o
s
ed
d
ataset
En
g
l
i
sh
H
i
n
d
i
O
d
i
a
Y
o
u
n
e
e
d
t
o
g
r
o
w
u
p
त
ु
बड
ह
ो
न
े
की
ज
र
त
है
(
T
u
m
h
e
a
n
b
a
ḍ
a
ā
h
o
n
e
k
ī
j
a
rū
r
a
t
h
a
i
)
ତ
ୁ
ମ
େ
ବଡ
ମ
େ
ବ
ା
ଆବ
ୟକ
(
T
u
m
e
b
a
d
a
h
e
b
a
a
b
a
sh
y
a
k
a
)
Y
o
u
n
e
e
d
t
o
h
e
l
p
m
e
त
ु
म
ेर
ी
मद
द
क
र
न
ी
ह
ो
ग
ी
(
T
u
m
h
e
a
n
m
e
r
ī
m
a
d
a
d
k
a
r
a
n
ī
h
o
g
ī
)
ତ
ୁ
ମ
େ
ମ
େ
ା
ମ
ତ
ସ
ା
େ
ା
ୟ
କର
ି
ବ
ା
ଆବ
ୟକ
(
T
u
m
e
m
o
t
e
sa
h
a
j
y
a
k
a
ri
b
a
a
b
a
s
h
y
a
k
a
)
Y
o
u
n
e
e
d
t
o
w
a
k
e
u
p
आप
क
ो
ज
ग
न
े
की
ज
र
त
है
(
Āp
a
k
o
j
ā
g
a
n
e
k
ī
j
a
r
ū
ra
t
h
a
i
)
ତ
ୁ
ମ
େ
ଜା
ତ
ମ
େ
ବ
ା
ଆ
ବ
ୟ
କ
(
T
u
m
e
j
a
g
ra
t
a
h
e
b
a
a
b
a
sh
y
a
k
a
)
Y
o
u
n
e
v
e
r
h
e
l
p
e
d
me
त
ु
मन
े
क
भ
ी
म
ेर
ी
मद
द
न
ह
ी
की
(
T
u
m
a
n
e
k
a
b
h
ī
m
e
rī
m
a
d
a
d
n
a
h
ī
a
n
k
ī
)
ତ
ୁ
ମ
େ
ମ
େ
ା
ମ
ତ
ମ
କମ
ବ
ସ
ା
େ
ା
ୟ
କର
ି
ନ
ା
େ
ଁ
(
T
u
m
e
m
o
t
e
k
e
b
e
s
a
h
a
j
y
a
k
a
ri
n
a
h
a
n
)
2
.
2
.
T
ra
ini
ng
da
t
a
s
et
Data
p
r
ep
ar
atio
n
is
a
f
u
n
d
a
m
en
tal
s
tep
in
an
y
m
ac
h
in
e
lear
n
in
g
task
,
co
m
p
r
is
in
g
clea
n
in
g
,
tr
an
s
f
o
r
m
in
g
,
an
d
o
r
g
a
n
izin
g
r
aw
d
ata.
T
o
tr
ain
HOBMT
m
o
d
el,
it
is
en
s
u
r
ed
th
at
Hin
d
i
an
d
Od
ia
d
ata
ar
e
p
r
o
p
er
l
y
p
r
e
-
p
r
o
ce
s
s
ed
(
to
k
en
ized
,
p
ad
d
ed
,
s
en
ten
ce
len
g
th
s
,
an
d
v
o
ca
b
u
lar
y
s
ize)
.
T
h
e
to
k
en
izer
o
b
ject
ca
n
co
n
v
er
t
all
th
e
wo
r
d
s
o
f
a
d
ataset
in
to
u
n
iq
u
e
n
u
m
er
ical
in
d
ex
.
Par
allel
s
en
ten
ce
p
air
o
f
Hin
d
i
-
Od
ia
a
n
d
Od
ia
-
Hin
d
i
ar
e
tab
u
lated
in
T
ab
le
2
.
T
h
e
to
tal
d
ataset
co
n
s
is
ts
o
f
6
0
,
0
0
0
p
a
r
allel
s
en
ten
ce
s
,
o
f
w
h
ich
7
2
%
(
4
3
,
2
0
0
)
is
allo
ca
ted
f
o
r
tr
ain
i
n
g
,
8
%
(
4
,
8
0
0
)
f
o
r
v
alid
atio
n
,
an
d
r
em
ai
n
in
g
2
0
%
(
1
2
,
0
0
0
)
a
r
e
u
s
ed
f
o
r
test
in
g
.
Hin
d
i
h
as
1
0
0
7
8
,
a
n
d
Od
ia
h
a
s
1
1
3
2
1
t
o
k
en
s
f
o
r
th
is
p
ar
ticu
lar
s
y
s
tem
.
Fig
u
r
e
1
co
m
p
ar
es
th
e
d
is
tr
ib
u
tio
n
o
f
a
p
ar
ticu
lar
f
ea
t
u
r
e
lik
e
s
en
ten
ce
len
g
th
,
o
f
th
e
Hin
d
i
an
d
Od
ia
d
atasets
b
y
d
is
p
lay
in
g
two
h
is
to
g
r
a
m
s
s
id
e
b
y
s
id
e.
Par
ticu
lar
ly
in
co
n
tr
ast
to
th
e
Hin
d
i
s
en
ten
ce
len
g
th
s
,
th
e
Od
ia
s
en
ten
ce
s
ar
e
in
m
ax
im
u
m
f
r
eq
u
en
cy
.
So
th
at
th
e
s
en
te
n
ce
len
g
th
is
tak
en
u
p
to
8
.
Sen
ten
ce
s
eq
u
e
n
ce
s
ar
e
co
n
v
er
te
d
in
to
in
teg
e
r
s
e
q
u
en
ce
s
.
T
o
m
ak
e
th
e
s
en
ten
ce
s
o
f
th
e
s
am
e
l
en
g
th
,
p
ad
d
in
g
is
ap
p
lied
s
o
th
at
z
er
o
s
ar
e
ad
d
e
d
to
th
e
en
d
o
f
th
e
s
eq
u
en
ce
.
T
ab
le
2
.
Data
s
et
s
p
lits
C
a
t
e
g
o
r
y
V
a
l
u
e
To
t
a
l
p
a
r
a
l
l
e
l
s
e
n
t
e
n
c
e
s
6
0
,
0
0
0
Tr
a
i
n
i
n
g
+
v
a
l
i
d
a
t
i
o
n
4
8
,
0
0
0
(
4
3
,
2
0
0
+
4
,
8
0
0
)
Te
st
i
n
g
1
2
,
0
0
0
H
i
n
d
i
v
o
c
a
b
u
l
a
r
y
si
z
e
1
0
0
7
8
O
d
i
a
v
o
c
a
b
u
l
a
r
y
si
z
e
1
1
3
2
1
Fig
u
r
e
1
.
His
to
g
r
am
o
f
Hin
d
i
an
d
Od
ia
d
ataset
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
2
,
J
u
n
e
2
0
2
6
:
6
8
7
-
695
690
2
.
3
.
Dee
p lea
rning
mo
dels
B
y
en
h
an
cin
g
n
e
u
r
al
n
etwo
r
k
ar
ch
itectu
r
e
,
m
a
n
y
c
o
m
p
le
x
p
r
o
b
lem
s
in
DL
ca
n
b
e
r
e
s
o
lv
ed
.
I
n
s
eq
2
s
eq
lear
n
in
g
,
th
e
R
NN
a
n
d
its
v
ar
iatio
n
s
ar
e
v
er
y
h
el
p
f
u
l.
T
h
e
m
o
s
t
p
o
p
u
lar
s
eq
2
s
eq
tech
n
iq
u
e
s,
lik
e
L
STM
an
d
GR
U
[
1
0
]
–
[
1
2
]
ar
e
d
escr
ib
ed
in
t
h
e
s
ec
tio
n
s
.
2.
3
.
1
.
Rec
urre
nt
neura
l net
wo
rk
s
T
h
e
s
h
o
r
tco
m
in
g
s
o
f
co
n
v
en
ti
o
n
al
n
eu
r
al
n
etwo
r
k
s
,
s
u
ch
as
f
ee
d
f
o
r
war
d
n
eu
r
al
n
etwo
r
k
s
(
FNNs
)
,
in
p
r
o
ce
s
s
in
g
s
eq
u
e
n
tial
d
ata
we
r
e
ad
d
r
ess
ed
b
y
t
h
e
in
tr
o
d
u
ct
io
n
o
f
R
NNs.
FNNs
ar
e
in
ef
f
icien
t
at
ca
p
tu
r
in
g
d
ep
en
d
e
n
cies b
ec
au
s
e
th
ey
p
r
o
ce
s
s
in
p
u
ts
in
d
ep
en
d
en
tly
,
d
i
s
r
eg
ar
d
in
g
co
n
tex
t a
n
d
o
r
d
er
.
B
y
ad
d
in
g
r
ec
u
r
r
e
n
t
co
n
n
ec
tio
n
s
,
R
NNs
g
et
ar
o
u
n
d
th
is
an
d
en
a
b
le
d
ata
to
m
o
v
e
b
etwe
en
tim
e
s
tep
s
.
I
n
o
r
d
e
r
to
lear
n
tem
p
o
r
al
d
ep
en
d
e
n
cies
an
d
m
a
n
ag
e
v
a
r
iab
le
-
len
g
th
in
p
u
ts
,
R
NNs
ca
n
r
etain
in
ter
n
al
m
em
o
r
y
,
wh
er
e
ea
ch
o
u
tp
u
t
is
f
ed
b
ac
k
as
in
p
u
t
to
th
e
s
u
b
s
eq
u
en
t
s
tep
[
1
2
]
,
[
2
4
]
.
R
NNs
p
r
i
m
ar
ily
u
s
e
r
ec
en
t
d
ata
wh
en
p
r
o
ce
s
s
in
g
len
g
th
y
s
en
ten
ce
s
b
ec
au
s
e
th
e
v
an
is
h
in
g
g
r
a
d
ien
t
p
r
o
b
lem
d
u
r
in
g
b
a
ck
p
r
o
p
ag
atio
n
ca
u
s
es
ea
r
lier
in
f
o
r
m
atio
n
to
f
ad
e
.
T
h
e
ex
p
lo
d
in
g
g
r
ad
ien
t
p
r
o
b
l
em
,
o
n
th
e
o
th
er
h
a
n
d
,
ca
n
al
s
o
r
esu
lt
f
r
o
m
g
r
a
d
ien
ts
g
r
o
w
in
g
ex
p
o
n
en
tial
ly
.
GR
Us
an
d
L
STM
s
wer
e
d
ev
elo
p
ed
to
o
v
e
r
co
m
e
th
ese
is
s
u
es,
en
h
an
ci
n
g
l
o
n
g
-
te
r
m
m
e
m
o
r
y
r
eten
tio
n
an
d
s
tab
ilizin
g
tr
ain
in
g
[
2
5
]
.
2.
3
.
2
.
L
o
ng
s
ho
rt
-
t
er
m m
e
mo
ry
T
h
is
m
o
d
el
wa
s
d
ev
elo
p
ed
i
n
1
9
9
7
to
i
d
en
tify
lo
n
g
-
ter
m
d
ep
en
d
en
cies
in
s
en
ten
ce
s
[
2
6
]
.
Usi
n
g
a
ce
ll
s
tate
an
d
th
r
ee
g
ates
,
“f
o
r
g
et,
u
p
d
ate,
an
d
o
u
t
p
u
t”
t
o
r
eg
u
late
in
f
o
r
m
atio
n
f
lo
w.
I
t
o
v
e
r
co
m
es
th
e
d
r
awb
ac
k
s
o
f
R
NNs.
T
h
ese
th
r
ee
g
ates
em
p
lo
y
s
ig
m
o
i
d
ac
tiv
atio
n
s
,
with
v
alu
es
r
an
g
in
g
f
r
o
m
0
to
1
,
r
etain
r
elev
an
t
d
ata
,
an
d
d
is
ca
r
d
th
e
r
est.
T
h
e
o
u
tp
u
t
g
ate
r
e
g
u
l
ates
th
e
s
u
b
s
eq
u
en
t
h
id
d
en
s
tate,
th
e
f
o
r
g
et
g
ate
ch
o
o
s
es wh
at
s
h
o
u
ld
b
e
k
e
p
t,
an
d
th
e
u
p
d
ate
g
ate
d
ec
id
es w
h
en
to
u
p
d
ate
th
e
m
em
o
r
y
[
2
7
]
.
2.
3
.
3
.
G
a
t
ed
re
curr
ent
un
it
GR
U
h
as
t
wo
g
ates
r
ath
er
th
an
th
r
ee
g
ates
an
d
f
ewe
r
p
ar
am
eter
s
,
wh
ich
s
im
p
lifie
s
L
STM
.
B
y
co
m
b
in
in
g
th
e
in
p
u
t
an
d
f
o
r
g
et
g
ates
in
to
a
s
in
g
le
“
u
p
d
ate
g
ate,
”
it
en
s
u
r
es
th
at
cr
u
cial
in
f
o
r
m
atio
n
is
p
r
eser
v
ed
o
v
er
tim
e.
W
h
ile
th
e
r
eset
g
ate
r
eg
u
lates
th
e
am
o
u
n
t
o
f
h
is
to
r
ical
d
ata
co
n
s
id
er
ed
,
th
e
u
p
d
ate
g
ate
estab
lis
h
es
th
e
f
in
al
ce
ll
s
tate.
Pre
v
io
u
s
d
ata
is
i
g
n
o
r
ed
if
t
h
e
r
eset
g
ate
is
clo
s
e
t
o
ze
r
o
an
d
in
f
o
r
m
atio
n
is
m
ain
tain
ed
if
th
e
u
p
d
ate
g
ate
is
clo
s
e
to
o
n
e
[
1
1
]
,
[
2
8
]
.
S
o
m
e
k
e
y
d
if
f
er
e
n
ce
s
h
a
v
e
b
e
en
f
o
u
n
d
f
r
o
m
th
r
ee
m
o
d
els
,
wh
ich
ar
e
d
is
cu
s
s
ed
as
f
o
llo
ws
:
i)
R
NN
s
ar
e
v
u
ln
er
ab
le
to
v
an
is
h
in
g
an
d
e
x
p
lo
d
i
n
g
g
r
ad
ie
n
ts
d
u
e
to
its
b
asic
s
tr
u
ctu
r
e
;
ii)
L
STM
s
em
p
lo
y
d
is
tin
ct
ce
ll
s
tates
an
d
g
ates
f
o
r
m
o
r
e
p
r
ec
i
s
e
co
n
tr
o
l
o
v
e
r
th
e
in
f
o
r
m
atio
n
f
l
o
w
;
an
d
iii)
G
R
U
u
s
e
s
f
e
w
e
r
p
a
r
a
m
e
te
r
s
a
n
d
i
s
a
n
a
l
t
e
r
n
a
ti
v
e
to
L
S
T
M
,
c
o
m
m
o
n
l
y
g
e
n
e
r
a
t
i
n
g
c
o
m
p
a
r
a
b
l
e
p
e
r
f
o
r
m
a
n
c
e
s
.
Pro
p
er
p
ar
am
ete
r
s
ettin
g
is
cr
u
cial
f
o
r
s
u
cc
ess
f
u
l
o
u
tc
o
m
es
b
ec
au
s
e
ad
ju
s
tin
g
th
e
v
alu
es
o
f
p
ar
am
eter
s
ca
n
af
f
ec
t
h
o
w
q
u
i
ck
ly
,
ac
c
u
r
ately
,
an
d
ef
f
icien
tly
th
e
p
r
o
g
r
am
r
u
n
s
.
So
m
e
co
m
m
o
n
f
ac
to
r
s
u
s
ed
to
d
ef
in
e
th
e
d
is
cu
s
s
ed
m
o
d
el
,
lik
e
s
ize
o
f
in
p
u
t
a
n
d
o
u
t
p
u
t
v
o
ca
b
u
lar
y
,
le
n
g
th
o
f
i
n
p
u
t
an
d
o
u
tp
u
t
s
eq
u
en
ce
s
as we
ll a
s
th
e
n
u
m
b
er
o
f
h
id
d
e
n
u
n
its
.
T
h
e
d
is
cu
s
s
ed
m
o
d
els
ar
e
co
m
p
iled
with
d
if
f
er
en
t p
a
r
am
eter
s
,
wh
ich
is
g
iv
en
in
T
a
b
le
3
.
T
ab
le
3
.
Par
am
eter
s
ettin
g
s
P
a
r
a
me
t
e
r
V
a
l
u
e
O
p
t
i
mi
z
e
r
R
o
o
t
me
a
n
sq
u
a
r
e
p
r
o
p
a
g
a
t
i
o
n
(
R
M
S
P
r
o
p
)
Lo
ss f
u
n
c
t
i
o
n
S
p
a
r
se
c
a
t
e
g
o
r
i
c
a
l
c
r
o
ss
e
n
t
r
o
p
y
Le
a
r
n
i
n
g
r
a
t
e
0
.
0
0
1
,
0
.
0
1
H
i
d
d
e
n
s
t
a
t
e
s
5
1
2
B
a
t
c
h
si
z
e
1
2
8
,
2
5
6
,
5
1
2
Ep
o
c
h
s
3
0
,
6
0
,
1
0
0
M
a
x
i
m
u
m
l
e
n
g
t
h
o
f
b
o
t
h
se
n
t
e
n
c
e
s
8
2
.4
.
E
v
a
lua
t
i
o
n m
et
rics
Fo
r
an
y
co
m
p
ar
ativ
e
s
tu
d
y
,
e
v
alu
atio
n
m
u
s
t
b
e
r
ap
id
,
p
r
ec
is
e,
ec
o
n
o
m
ical,
an
d
tim
e
-
ef
f
i
cien
t.
T
h
e
s
y
s
tem
is
tr
ain
ed
u
s
in
g
th
e
s
a
m
e
co
r
p
u
s
o
f
lan
g
u
ag
e
p
air
s
to
g
u
ar
an
tee
th
at
th
e
e
v
alu
atio
n
is
co
n
s
is
ten
t
ac
r
o
s
s
all
m
o
d
els.
I
n
th
is
ca
s
e,
B
L
E
U
[
1
5
]
,
ME
T
E
OR
[
1
6
]
,
W
E
R
[
1
7
]
,
an
d
C
E
R
[
1
8
]
s
co
r
es
h
av
e
b
ee
n
u
s
ed
to
g
u
id
e
th
e
e
v
alu
atio
n
a
n
d
c
o
m
p
ar
is
o
n
o
f
th
e
s
y
s
tem
s
.
B
L
E
U
m
etr
ic
m
ea
s
u
r
es
th
e
s
im
ilar
ity
b
etwe
en
a
m
ac
h
i
n
e
-
t
r
an
s
lated
tex
t
to
a
r
ef
er
en
ce
t
r
an
s
latio
n
.
T
h
is
s
co
r
e
r
an
g
e
lies
b
etwe
en
0
to
1
.
W
h
er
e
0
in
d
icate
s
n
o
m
atch
an
d
1
in
d
icate
s
p
er
f
ec
t
m
atch
b
etwe
en
th
e
p
r
ed
icted
s
en
te
n
ce
an
d
r
e
f
er
e
n
ce
s
en
ten
ce
.
I
t
is
m
o
r
e
c
o
m
p
letely
d
ep
e
n
d
en
t
o
n
p
r
ec
is
io
n
.
I
t
ig
n
o
r
es
s
im
ilar
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
A
s
s
es
s
men
t o
f d
ee
p
lea
r
n
in
g
b
a
s
ed
Hin
d
i O
d
ia
b
id
ir
ec
tio
n
a
l
ma
ch
in
e
tr
a
n
s
la
tio
n
…
(
S
u
b
h
a
s
h
r
ee
S
a
tp
a
th
y)
691
wo
r
d
s
an
d
s
ea
r
ch
es
f
o
r
ex
ac
t
wo
r
d
s
.
T
h
e
ME
T
E
OR
s
co
r
e
co
m
p
u
tes
th
e
h
ar
m
o
n
ic
m
e
an
u
s
in
g
weig
h
ted
p
r
ec
is
io
n
an
d
r
ec
all.
Sin
ce
it
co
n
ce
n
tr
ates
o
n
co
r
r
elatio
n
with
h
u
m
an
tr
an
s
latio
n
s
at
t
h
e
s
eg
m
en
t
lev
el,
wh
er
ea
s
th
e
B
L
E
U
s
co
r
e
f
o
c
u
s
es
o
n
co
r
r
elatio
n
at
th
e
c
o
r
p
u
s
lev
el,
th
is
s
co
r
e
r
eso
lv
es
a
n
u
m
b
er
o
f
is
s
u
es
with
th
e
B
L
E
U
s
co
r
e.
W
E
R
co
m
p
u
tes
th
e
r
atio
o
f
er
r
o
r
s
o
n
s
u
b
s
titu
tio
n
s
,
in
s
er
tio
n
s
,
an
d
d
eletio
n
s
in
th
e
s
y
s
tem
'
s
o
u
tp
u
t
to
th
e
to
tal
n
u
m
b
er
o
f
wo
r
d
s
in
th
e
r
ef
e
r
e
n
ce
tex
t.
C
E
R
ca
lcu
lates
th
e
d
is
tan
ce
b
etwe
en
th
e
m
ac
h
in
e
-
tr
an
s
lated
te
x
t
,
wh
ic
h
is
k
n
o
wn
as
t
h
e
ca
n
d
id
ate
te
x
t
an
d
th
e
co
r
r
ec
t,
h
u
m
an
-
tr
a
n
s
lated
tex
t
,
wh
ich
is
k
n
o
wn
as th
e
r
ef
er
en
ce
tex
t a
t th
e
ch
ar
ac
ter
lev
el.
Fo
r
tr
ain
in
g
p
u
r
p
o
s
e
s
,
g
r
ap
h
ics p
r
o
ce
s
s
in
g
u
n
it (
GPU
)
is
an
ex
ce
llen
t
s
u
b
s
titu
te
f
o
r
t
h
e
ce
n
tr
al
p
r
o
ce
s
s
in
g
u
n
it
(
C
PU
)
in
ter
m
s
o
f
h
an
d
lin
g
an
d
s
p
ee
d
y
co
m
p
u
tatio
n
.
I
t
s
o
lv
es
th
e
p
r
o
b
lem
o
f
co
m
p
l
ex
co
m
p
u
tatio
n
an
d
lead
s
th
e
s
y
s
tem
to
war
d
s
g
o
o
d
p
er
f
o
r
m
an
ce
,
s
u
p
p
o
r
tin
g
m
ass
iv
e
p
ar
allel
co
m
p
u
tatio
n
.
Alm
o
s
t
o
n
e
e
p
o
ch
tak
es
less
th
an
o
n
e
m
i
n
u
te
u
s
in
g
th
e
s
y
s
tem
co
n
f
ig
u
r
atio
n
d
ep
icted
in
T
a
b
le
4
.
T
ab
le
4
.
Sy
s
tem
co
n
f
ig
u
r
atio
n
C
o
m
p
o
n
e
n
t
S
p
e
c
i
f
i
c
a
t
i
o
n
P
y
t
h
o
n
3
G
o
o
g
l
e
C
o
mp
u
t
e
E
n
g
i
n
e
b
a
c
k
e
n
d
(
G
P
U
)
T4
G
P
U
S
y
st
e
m
R
A
M
1
2
.
7
G
B
G
P
U
R
A
M
1
5
.
0
G
B
D
i
sk
1
1
2
.
6
G
B
P
r
o
c
e
ss
o
r
I
n
t
e
l
C
o
r
e
i
5
2
.
5
.
Wo
r
k
f
lo
w
dia
g
ra
m
T
h
e
ch
o
ice
o
f
s
elec
tin
g
b
etwe
en
th
ese
m
o
d
els
is
p
r
im
ar
ily
b
ased
o
n
th
e
n
ee
d
o
f
task
in
c
lu
d
in
g
t
h
e
co
m
p
u
tatio
n
al
r
eso
u
r
ce
s
th
at
ar
e
av
ailab
le,
th
e
co
m
p
le
x
ity
o
f
th
e
d
ep
en
d
en
cies
,
a
n
d
t
h
e
s
eq
u
en
ce
len
g
th
.
Seq
u
en
tial
m
o
d
els
ar
e
s
im
p
le
m
eth
o
d
o
f
b
u
ild
in
g
n
e
u
r
al
n
et
wo
r
k
s
b
y
s
tack
in
g
lay
er
s
lin
ea
r
ly
,
an
d
th
e
o
u
tp
u
t
o
f
ea
ch
la
y
er
f
ee
d
in
g
d
ir
ec
tly
in
to
th
e
n
e
x
t
lay
er
[
2
9
]
.
Fig
u
r
e
2
r
ep
r
esen
ts
th
e
wo
r
k
f
lo
w
to
estab
lis
h
HOBMT
s
y
s
tem
,
in
clu
d
in
g
s
ev
er
al
d
is
tin
ct
s
t
ep
s
lik
e
p
ar
am
eter
s
etti
n
g
,
tr
ain
in
g
d
ataset
,
an
d
m
o
d
el
ev
alu
atio
n
,
an
d
f
in
ally
to
s
av
e
th
e
b
est m
o
d
els
.
Fig
u
r
e
3
is
an
illu
s
tr
atio
n
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
els
u
s
in
g
an
en
co
d
er
-
d
ec
o
d
er
m
ec
h
a
n
is
m
.
T
h
e
m
o
d
e
l
r
ec
eiv
es
in
p
u
t
s
eq
u
en
ce
s
with
o
n
e
wo
r
d
f
o
r
ea
c
h
tim
e
s
tep
.
T
o
m
atch
th
e
v
o
ca
b
u
la
r
y
in
th
e
tar
g
et
d
ataset,
ea
ch
wo
r
d
is
en
co
d
e
d
as
a
d
is
tin
ct
in
teg
er
.
E
v
e
r
y
wo
r
d
is
co
n
v
er
ted
in
to
a
v
ec
to
r
u
s
in
g
em
b
ed
d
in
g
an
d
th
e
v
ec
to
r
s
ize
d
ep
en
d
s
o
n
th
e
lev
el
o
f
v
o
ca
b
u
lar
y
c
o
m
p
lex
ity
.
I
n
en
co
d
e
r
lay
er
,
t
h
e
cu
r
r
e
n
t
w
o
r
d
v
ec
to
r
is
g
iv
en
th
e
co
n
tex
t
f
r
o
m
wo
r
d
v
ec
to
r
s
in
ea
r
lier
tim
e
s
tep
s
.
Den
s
e
lay
er
,
wh
er
e
th
e
en
co
d
ed
in
p
u
t
i
s
d
ec
o
d
ed
in
to
t
h
e
ap
p
r
o
p
r
iate
tr
an
s
latio
n
s
eq
u
en
ce
u
s
in
g
th
ese
f
u
lly
c
o
n
n
ec
te
d
lay
er
s
.
T
h
e
o
u
t
p
u
ts
ca
n
b
e
m
ap
p
ed
t
o
th
e
tar
g
et
d
ataset
v
o
ca
b
u
lar
y
af
ter
b
ei
n
g
r
etu
r
n
ed
as
a
s
er
ies
o
f
in
teg
er
s
o
r
en
c
o
d
ed
v
ec
to
r
s
[
2
5
]
,
[
3
0
]
.
T
h
e
R
ep
ea
tVec
to
r
lay
er
r
ep
ea
ts
th
i
s
co
n
tex
t
v
ec
to
r
o
v
er
t
im
e
,
f
e
ed
in
g
it
to
th
e
d
ec
o
d
er
.
T
h
en
th
e
s
ec
o
n
d
lay
e
r
o
f
d
is
cu
s
s
ed
m
o
d
els
ac
ts
a
s
th
e
d
ec
o
d
er
.
T
h
e
d
en
s
e
lay
er
with
S
o
f
tMa
x
ac
tiv
atio
n
o
u
tp
u
ts
th
e
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
tio
n
o
v
er
th
e
o
u
tp
u
t
v
o
ca
b
u
lar
y
[
3
1
]
.
Fig
u
r
e
2
.
HOBMT
s
y
s
tem
m
o
d
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
2
,
J
u
n
e
2
0
2
6
:
6
8
7
-
695
692
Fig
u
r
e
3
.
Pro
p
o
s
ed
MT
s
y
s
te
m
u
s
in
g
an
e
n
co
d
e
r
-
d
ec
o
d
er
m
ec
h
an
is
m
Ma
th
em
atica
lly
,
it
tak
es
th
e
cu
r
r
en
t
in
p
u
t
(
)
an
d
th
e
m
em
o
r
y
o
f
th
e
p
ast
o
r
th
e
p
r
ev
i
o
u
s
h
id
d
en
s
tate
ℎ
(
−
1
)
to
ca
lcu
late
a
n
ew
m
em
o
r
y
o
r
n
ew
h
id
d
e
n
s
tate
(
ℎ
)
.
T
h
en
th
e
n
ew
m
em
o
r
y
(
ℎ
)
is
u
s
ed
to
p
r
o
d
u
ce
an
o
u
tp
u
t
(
)
.
T
h
e
weig
h
t m
atr
ix
ℎℎ
is
c
o
n
n
ec
tin
g
f
r
o
m
p
r
ev
io
u
s
h
id
d
en
s
tate
to
th
e
cu
r
r
en
t h
id
d
e
n
s
tate
an
d
ℎ
is
co
n
n
ec
tin
g
t
h
e
cu
r
r
en
t
in
p
u
t
to
th
e
cu
r
r
en
t
h
id
d
en
s
tate.
(
ℎ
)
is
k
n
o
w
n
as
th
e
b
ias
v
ec
to
r
o
f
h
id
d
en
s
tates.
ℎ
is
t
h
e
h
y
p
er
b
o
li
c
ta
n
g
e
n
t
a
cti
v
a
ti
o
n
f
u
n
cti
o
n
,
wh
i
ch
s
q
u
as
h
es
t
h
e
v
al
u
es
b
et
wee
n
-
1
a
n
d
1
.
T
h
e
(
1
)
an
d
(
2
)
ca
n
b
e
u
s
ed
to
r
ep
r
esen
t th
e
f
u
n
d
a
m
en
tal
s
tr
u
ctu
r
e
o
f
a
n
R
NN.
ℎ
=
ℎ
(
ℎℎ
ℎ
−
1
+
ℎ
+
ℎ
)
(
1
)
T
h
e
o
u
tp
u
t
(
)
at
th
e
cu
r
r
en
t
ti
m
e
s
tep
“t”
is
k
n
o
wn
as
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
tio
n
o
r
p
r
ed
icti
o
n
.
T
h
e
weig
h
t
m
atr
ix
(
ℎ
)
is
m
atr
ix
co
n
n
ec
tin
g
t
h
e
h
id
d
en
s
tate
to
th
e
o
u
tp
u
t
an
d
th
e
o
u
tp
u
t
b
ias
v
ec
to
r
is
(
)
wh
ich
is
ca
lcu
lated
as
in
(
2
)
.
=
ℎ
ℎ
+
(
2
)
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
R
eliab
ilit
y
,
r
o
b
u
s
tn
ess
,
an
d
e
f
f
ec
tiv
en
ess
o
n
a
v
ar
iety
o
f
task
s
ar
e
u
s
ed
to
ev
alu
ate
t
h
e
o
v
er
all
p
er
f
o
r
m
an
ce
o
f
th
e
MT
s
y
s
te
m
.
A
d
etailed
e
v
alu
atio
n
n
o
t
o
n
ly
h
elp
s
m
o
n
ito
r
a
m
o
d
el'
s
p
o
ten
tial
b
u
t
also
id
en
tifie
s
ar
ea
s
th
at
n
ee
d
s
o
m
e
wo
r
k
.
Fig
u
r
e
4
s
ig
n
if
ies
th
e
3
0
-
e
p
o
ch
r
esu
lts
o
f
HOBMT
s
y
s
tem
.
T
y
p
ical
ly
,
B
L
E
U
s
co
r
e
i
s
ac
ce
p
ted
to
a
u
to
m
atica
lly
ev
alu
ate
th
e
q
u
a
lity
o
f
tr
an
s
latio
n
.
Sco
r
e
v
alu
es
ar
e
o
b
s
er
v
ed
at
30
-
ep
o
ch
s
.
T
h
is
in
d
icate
s
th
at
th
e
s
im
p
le
R
NN’
s
B
L
E
U
s
co
r
e
o
u
tp
er
f
o
r
m
s
L
STM
an
d
GR
U
f
o
r
th
e
HO
B
MT
tr
an
s
latio
n
task
s
f
o
r
th
e
cu
r
r
e
n
t
d
ataset
an
d
t
r
ain
in
g
s
etu
p
.
So
m
e
an
aly
s
is
an
d
k
ey
p
r
o
b
lem
s
wh
ich
is
d
ef
in
ed
th
at
R
NN
h
as a
ch
iev
ed
g
o
o
d
r
esu
lt
s
in
co
m
p
ar
is
o
n
to
L
STM
an
d
GR
U.
Fig
u
r
e
4
.
Per
f
o
r
m
an
c
e
co
m
p
ar
is
o
n
o
f
HOBMT
s
y
s
tem
s
3
.
1
.
Sente
nce
leng
t
hs
T
h
e
s
en
ten
ce
le
n
g
th
s
ar
e
s
m
all
an
d
s
im
p
le
with
in
th
is
ex
p
er
im
en
t.
R
NN
g
en
er
ates
tex
t
s
e
q
u
en
tially
b
y
p
r
ed
ictin
g
th
e
n
ex
t
c
h
ar
ac
t
er
o
r
wo
r
d
b
ased
o
n
th
eir
p
r
io
r
s
eq
u
en
ce
s
.
L
STM
an
d
GR
U
ce
lls
ar
e
d
esig
n
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
A
s
s
es
s
men
t o
f d
ee
p
lea
r
n
in
g
b
a
s
ed
Hin
d
i O
d
ia
b
id
ir
ec
tio
n
a
l
ma
ch
in
e
tr
a
n
s
la
tio
n
…
(
S
u
b
h
a
s
h
r
ee
S
a
tp
a
th
y)
693
to
ca
p
tu
r
e
l
o
n
g
-
r
an
g
e
d
ep
e
n
d
e
n
cies
in
s
eq
u
en
ce
s
,
wh
ic
h
is
n
o
t
s
u
itab
le
f
o
r
s
h
o
r
t
a
n
d
s
im
p
l
e
s
en
ten
ce
s
.
R
NN
ap
p
r
o
ac
h
h
as
g
iv
en
r
is
e
to
a
n
ew
o
p
tim
is
m
,
p
r
o
d
u
cin
g
b
ett
er
r
esu
lt
s
f
o
r
s
h
o
r
ter
s
en
ten
ce
s
in
s
tead
o
f
lo
n
g
er
one
s
.
E
v
alu
atio
n
m
etr
ics s
co
r
e
o
f
R
NN
n
etwo
r
k
is
m
o
r
e
th
an
4
0
% a
s
co
m
p
ar
e
to
L
STM
an
d
GR
U.
3
.
2
.
Re
m
em
bering
a
nd
re
a
s
o
nin
g
T
h
e
co
n
tin
u
o
u
s
v
ec
to
r
r
ep
r
esen
tatio
n
f
o
r
all
th
ese
th
r
ee
n
etwo
r
k
s
is
in
ca
p
ab
le
to
r
ec
all
th
e
en
tire
s
o
u
r
ce
s
en
ten
ce
.
Ob
tain
i
n
g
th
e
co
r
r
ec
t
tar
g
et
tr
an
s
latio
n
th
r
o
u
g
h
d
ec
o
d
in
g
f
r
o
m
t
h
is
r
ep
r
esen
tatio
n
is
q
u
ite
ch
allen
g
in
g
.
Ad
d
i
tio
n
ally
,
M
T
is
a
m
o
r
e
co
m
p
lex
p
r
o
b
le
m
th
at
ca
lls
f
o
r
r
ich
r
ea
s
o
n
in
g
o
p
er
atio
n
s
,
u
n
lik
e
o
th
er
s
eq
2
s
eq
NL
P task
s
.
3
.
3
.
Co
m
pu
t
a
t
io
na
l c
o
m
plex
it
y
Dee
p
n
eu
r
al
n
etwo
r
k
m
o
d
el
t
r
ain
in
g
is
a
tim
e
-
co
n
s
u
m
i
n
g
ta
s
k
.
I
t
s
p
ec
if
ies
th
e
m
ath
em
ati
ca
l
g
r
o
wth
r
ate
o
f
a
p
ar
ticu
lar
alg
o
r
ith
m
i
c
r
eso
u
r
ce
.
T
r
ain
in
g
n
e
u
r
al
m
ac
h
in
e
tr
a
n
s
latio
n
(
NM
T
)
s
y
s
tem
ca
n
tak
e
m
u
c
h
tim
e
o
r
m
em
o
r
y
,
e
v
en
with
p
o
wer
f
u
l G
PUs
.
3
.
4
.
E
rr
o
r
a
na
ly
s
is
T
h
is
an
aly
s
is
r
ev
ea
ls
th
e
s
p
ec
if
ic
lin
g
u
is
tic
p
atter
n
s
th
at
th
e
m
o
d
el
is
n
o
t
ca
p
tu
r
in
g
an
d
th
e
r
ea
s
o
n
f
o
r
t
h
e
l
o
w
s
co
r
e.
I
t
is
ch
allen
g
in
g
to
c
o
m
p
r
e
h
en
d
wh
y
th
es
e
NM
T
s
im
p
r
o
v
e
tr
an
s
latio
n
p
er
f
o
r
m
a
n
ce
o
r
wh
y
it
f
ails
b
ec
au
s
e
it
wo
r
k
s
wit
h
v
ar
ia
b
les
in
th
e
r
eal
-
v
alu
ed
co
n
tin
u
o
u
s
s
p
ac
e.
E
r
r
o
r
an
a
ly
s
is
co
n
tr
ib
u
tes
to
b
etter
m
o
d
el
p
er
f
o
r
m
a
n
ce
b
y
h
ig
h
lig
h
tin
g
f
laws,
m
o
d
el
tu
n
in
g
,
b
o
o
s
tin
g
o
v
er
all
ac
cu
r
a
cy
,
an
d
ad
d
r
ess
in
g
d
ata
clea
n
in
g
.
T
est
ca
s
e
r
esu
lts
ar
e
o
b
s
er
v
ed
at
3
0
th
ep
o
ch
o
f
th
is
d
ataset
i
s
d
is
p
lay
e
d
in
T
ab
le
5
.
On
e
m
is
tr
an
s
latio
n
ex
am
p
le
f
r
o
m
T
ab
le
5
is
a
ctu
al
-
ମ
ଯ
ି
ବ
ା
କ
ଚ
ି
ା
କର
ଛ
ି
(
mu
n
jib
a
ku
cin
ta
ka
r
u
ch
i
)
=
p
r
ed
icted
-
ମ
ଯ
ି
ବ
ା
କ
(
mu
jib
a
k
u
)
,
wh
er
e
t
h
e
s
o
u
r
ce
a
n
d
ta
r
g
et
wo
r
d
o
r
d
er
s
ar
e
in
c
o
r
r
ec
t.
Oth
er
c
r
iter
ia
f
o
r
ch
an
g
in
g
p
ar
am
eter
s
ar
e
ta
k
en
in
t
o
co
n
s
id
er
atio
n
to
d
eter
m
in
e
wh
et
h
er
im
p
r
o
v
em
en
t
is
n
ec
ess
ar
y
l
ik
e
(
n
o
o
f
ep
o
c
h
s
,
lear
n
in
g
r
ate,
b
atch
s
ize)
.
T
h
e
o
p
tim
u
m
p
er
f
o
r
m
a
n
ce
is
o
b
s
er
v
ed
,
with
ep
o
c
h
s
=3
0
,
lear
n
in
g
r
ate
=0
.
0
1
,
b
atch
s
ize
=1
2
8
.
T
h
e
s
o
u
r
ce
a
n
d
tar
g
et
test
ca
s
es wo
r
d
s
m
atch
in
g
o
f
R
NN
i
s
eq
u
al
as c
o
m
p
ar
e
t
o
L
STM
an
d
GR
U.
On
e
o
f
t
h
e
r
ea
s
o
n
s
f
o
r
I
n
d
ia'
s
lo
w
lev
el
o
f
d
ig
ital
liter
ac
y
is
lan
g
u
ag
e
b
ar
r
ie
r
s
.
T
h
is
w
o
r
k
ca
n
b
e
u
s
ed
to
cr
ea
te
web
co
n
ten
t
tr
an
s
latio
n
ap
p
licatio
n
s
,
wh
ich
will
les
s
en
th
e
lan
g
u
ag
e
b
ar
r
ier
.
I
t
wo
u
ld
en
co
u
r
a
g
e
m
o
r
e
p
e
o
p
le
to
u
s
e
d
ig
ital
p
latf
o
r
m
s
,
g
o
v
er
n
m
en
t
p
o
r
tals
,
an
d
ed
u
ca
tio
n
al
ap
p
s
,
wh
ich
wo
u
ld
aid
th
e
g
o
v
er
n
m
en
t'
s
D
ig
ital
I
n
d
ia
in
itiativ
e.
T
ec
h
n
iq
u
es
lik
e
atten
tio
n
m
ec
h
an
is
m
s
,
lar
g
e
lan
g
u
ag
e
m
o
d
els,
r
eg
u
lar
izatio
n
tech
n
iq
u
es,
an
d
tr
an
s
f
o
r
m
e
r
-
b
ased
m
o
d
els
lik
e
b
id
ir
ec
tio
n
al
en
co
d
er
r
e
p
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
er
s
(
B
E
R
T
)
,
g
e
n
er
ativ
e
p
r
e
-
tr
ain
e
d
tr
an
s
f
o
r
m
e
r
(
GPT)
,
an
d
m
u
ltil
in
g
u
al
b
id
ir
ec
tio
n
al
a
n
d
au
to
-
r
eg
r
ess
iv
e
tr
an
s
f
o
r
m
e
r
s
(
MBAR
T
)
m
ay
also
in
clu
d
e
a
s
p
ar
t
o
f
th
e
f
u
tu
r
e
s
co
p
e
o
f
th
is
wo
r
k
.
B
esid
es
th
is
,
p
r
o
ce
s
s
es
to
im
p
r
o
v
e
q
u
ality
ch
ec
k
in
g
,
as
well
as
u
s
in
g
d
ata
g
en
er
atio
n
f
o
r
p
ar
alle
l
co
r
p
o
r
a
s
tr
ateg
ies
,
ca
n
b
e
b
u
ilt to
f
u
r
th
er
u
p
lift
tr
an
s
latio
n
ef
f
icien
cy
.
T
ab
le
5
.
T
es
t c
ases
r
esu
lts
o
f
HOBMT
s
y
s
tem
s
T
y
p
e
HI
-
OD
OD
-
HI
R
N
N
LSTM
G
R
U
R
N
N
LSTM
G
R
U
A
c
t
u
a
l
वह
म
ह
(
v
a
h
m
a
i
a
n
h
ū
a
n
)
वह
म
ह
(
v
a
h
m
a
i
a
n
h
ū
a
n
)
वह
म
ह
(
v
a
h
m
a
i
a
n
h
ū
a
n
)
ମ
ଯ
ି
ବ
ା
କ
ଚ
ି
ା
କର
ଛ
ି
(
m
u
n
j
i
b
a
k
u
c
i
n
t
a
k
a
r
u
c
h
i
)
ମ
ଯ
ି
ବ
ା
କ
ଚ
ି
ା
କର
ଛ
ି
(
m
u
n
j
i
b
a
k
u
c
i
n
t
a
k
a
r
u
c
h
i
)
ମ
ଯ
ି
ବ
ା
କ
ଚ
ି
ା
କର
ଛ
ି
(
m
u
n
j
i
b
a
k
u
c
i
n
t
a
k
a
r
u
c
h
i
)
P
r
e
d
i
c
t
e
d
वह
म
ह
(
v
a
h
m
a
i
a
n
h
ū
a
n
)
वह
हो
है
(
v
a
h
h
o
h
a
i
)
वह
ह
(
v
a
h
h
ū
a
n
)
ମ
ଯ
ି
ବ
ା
କ
ଚ
ି
ା
କର
ଛ
ି
(
m
u
n
j
i
b
a
k
u
c
i
n
t
a
k
a
r
u
c
h
i
)
ମ
ଯ
ି
ବ
ା
କ
(
m
u
j
i
b
a
k
u
)
ମ
ଯ
ି
ବ
ା
କ
କର
ି
ବ
ି
(
m
u
n
j
i
b
a
k
u
k
a
r
i
b
i
)
4.
CO
NCLU
SI
O
N
Stan
d
ar
d
d
ata
f
o
r
I
n
d
ian
lan
g
u
ag
es
is
in
s
u
f
f
icien
t
an
d
n
o
t
d
ev
elo
p
e
d
to
im
p
lem
en
t
o
n
s
ev
er
al
DL
m
eth
o
d
s
ef
f
ec
tiv
el
y
.
L
ess
wo
r
k
h
as
b
ee
n
ac
co
u
n
ted
f
o
r
Hin
d
i
to
Od
ia
o
r
Od
ia
Hin
d
i
MT
s
y
s
tem
,
wh
ich
is
a
ch
allen
g
e
to
cr
ea
te
o
n
e
HOBMT
s
y
s
tem
.
I
t
em
p
h
asizes
o
n
b
id
ir
ec
tio
n
al
ev
al
u
atio
n
o
n
lo
w
-
r
eso
u
r
ce
I
n
d
ic
lan
g
u
ag
es.
T
h
ese
d
is
cu
s
s
ed
s
eq
2
s
eq
m
o
d
els
p
er
f
o
r
m
s
d
ec
en
t
at
th
e
s
am
e
tim
e
,
it
co
n
f
u
s
es
to
g
r
asp
th
e
im
p
o
r
tan
t
wo
r
d
s
in
s
ev
er
al
ca
s
es.
T
o
d
eter
m
in
e
th
e
b
est
m
o
d
el
p
er
f
o
r
m
s
b
est
f
o
r
MT
,
t
h
is
wo
r
k
co
m
p
ar
es
p
er
f
o
r
m
an
ce
s
o
f
v
ar
i
o
u
s
m
o
d
els
u
s
in
g
d
if
f
er
e
n
t
ev
al
u
atio
n
m
etr
ics.
T
h
is
an
aly
s
is
allo
ws
u
s
to
v
ali
d
at
e
an
d
co
m
p
ar
e
n
o
t
o
n
ly
th
e
ac
cu
r
a
cy
b
u
t
also
th
e
f
in
al
p
e
r
f
o
r
m
an
ce
o
f
t
h
e
m
o
d
els
with
r
esp
ec
t
to
th
eir
d
ataset,
tr
ain
in
g
tim
e,
an
d
p
ar
am
ete
r
s
.
I
n
co
n
tr
ast
to
L
STM
an
d
G
R
U
n
etwo
r
k
s
,
th
e
c
o
m
p
ar
is
o
n
d
em
o
n
s
tr
ates
th
at
R
NN
o
u
tp
er
f
o
r
m
s
th
em
o
n
s
h
o
r
ter
s
en
ten
ce
s
with
less
co
m
p
u
tatio
n
tim
e.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
e
au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
is
in
v
o
lv
ed
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
2
,
J
u
n
e
2
0
2
6
:
6
8
7
-
695
694
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Su
b
h
ash
r
ee
Satp
ath
y
✓
✓
✓
✓
✓
✓
✓
✓
Sm
itap
r
av
a
Mish
r
a
✓
✓
✓
✓
✓
✓
Ajit Ku
m
ar
Nay
ak
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
th
o
r
s
tate
s
n
o
co
n
f
lict o
f
in
ter
est
.
DATA AV
AI
L
AB
I
L
I
T
Y
Data
co
llected
f
r
o
m
th
e
o
p
e
n
s
o
u
r
ce
at
h
ttp
s
://www.
m
an
y
th
in
g
s
.
o
r
g
/a
n
k
i/
,
r
e
f
er
en
ce
n
u
m
b
er
[
2
3
]
.
Af
ter
th
at
m
an
u
ally
co
n
v
er
ted
in
to
Hin
d
i
-
Od
ia
b
ilin
g
u
al
f
o
r
m
.
RE
F
E
R
E
NC
E
S
[
1
]
S
.
B
.
D
a
s,
D
.
P
a
n
d
a
,
T.
K
.
M
i
s
h
r
a
,
a
n
d
B
.
K
.
P
a
t
r
a
,
“
S
t
a
t
i
s
t
i
c
a
l
m
a
c
h
i
n
e
t
r
a
n
sl
a
t
i
o
n
f
o
r
I
n
d
i
c
l
a
n
g
u
a
g
e
s
,
”
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pro
c
e
ssi
n
g
,
v
o
l
.
3
1
,
n
o
.
2
,
p
p
.
3
2
8
–
3
4
5
,
M
a
r
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
7
/
n
l
p
.
2
0
2
4
.
2
6
.
[
2
]
S
i
t
e
n
d
e
r
,
S
.
B
a
w
a
,
M
.
K
u
mar,
a
n
d
S
a
n
g
e
e
t
a
,
“
A
c
o
m
p
r
e
h
e
n
si
v
e
s
u
r
v
e
y
o
n
mac
h
i
n
e
t
r
a
n
sl
a
t
i
o
n
f
o
r
E
n
g
l
i
sh
,
H
i
n
d
i
a
n
d
S
a
n
s
k
r
i
t
l
a
n
g
u
a
g
e
s
,
”
J
o
u
r
n
a
l
o
f
Am
b
i
e
n
t
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
H
u
m
a
n
i
z
e
d
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
4
,
n
o
.
4
,
p
p
.
3
4
4
1
–
3
4
7
4
,
A
p
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
2
6
5
2
-
021
-
0
3
4
7
9
-
0.
[
3
]
B
.
S
.
H
a
r
i
sh
a
n
d
R
.
K
.
R
a
n
g
a
n
,
“
A
c
o
mp
r
e
h
e
n
s
i
v
e
s
u
r
v
e
y
o
n
I
n
d
i
a
n
r
e
g
i
o
n
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ssi
n
g
,
”
S
N
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
2
,
n
o
.
7
,
J
u
l
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
2
4
5
2
-
0
2
0
-
2
9
8
3
-
x.
[
4
]
A
.
K
a
l
y
a
n
i
a
n
d
P
.
S
.
S
a
j
j
a
,
“
A
r
e
v
i
e
w
o
f
m
a
c
h
i
n
e
t
r
a
n
sl
a
t
i
o
n
sy
s
t
e
ms
i
n
I
n
d
i
a
a
n
d
d
i
f
f
e
r
e
n
t
t
r
a
n
sl
a
t
i
o
n
e
v
a
l
u
a
t
i
o
n
met
h
o
d
o
l
o
g
i
e
s
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
2
1
,
n
o
.
2
3
,
p
p
.
1
6
–
2
3
,
Ju
l
.
2
0
1
5
,
d
o
i
:
1
0
.
5
1
2
0
/
2
1
8
4
0
-
4
9
1
7
.
[
5
]
O
.
B
o
j
a
r
e
t
a
l
.
,
“
F
i
n
d
i
n
g
s
o
f
t
h
e
2
0
1
5
w
o
r
k
s
h
o
p
o
n
s
t
a
t
i
st
i
c
a
l
ma
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
T
e
n
t
h
W
o
rks
h
o
p
o
n
S
t
a
t
i
st
i
c
a
l
M
a
c
h
i
n
e
T
ra
n
s
l
a
t
i
o
n
,
S
t
r
o
u
d
sb
u
r
g
,
U
S
A
,
2
0
1
5
,
p
p
.
1
–
4
6
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
W
1
5
-
3
0
0
1
.
[
6
]
N
.
Te
l
r
a
n
d
h
e
a
n
d
R
.
K
a
d
u
,
“
M
a
c
h
i
n
e
t
r
a
n
sl
a
t
i
o
n
mo
d
e
l
s
f
o
r
l
o
w
-
r
e
so
u
r
c
e
I
ndo
-
A
r
y
a
n
l
a
n
g
u
a
g
e
s
:
a
c
o
mp
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
o
f
t
e
c
h
n
i
q
u
e
s,
c
h
a
l
l
e
n
g
e
s
a
n
d
f
u
t
u
r
e
sc
o
p
e
,
”
i
n
2
0
2
5
3
r
d
D
MIH
ER
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
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
i
n
H
e
a
l
t
h
c
a
re
,
Ed
u
c
a
t
i
o
n
a
n
d
I
n
d
u
s
t
ry
(
I
D
I
C
AI
H
EI
)
,
N
o
v
.
2
0
2
5
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
D
I
C
A
I
H
E
I
6
5
9
9
1
.
2
0
2
5
.
1
1
3
7
8
2
1
4
.
[
7
]
J.
R
a
u
t
a
r
a
y
a
n
d
P
.
M
i
sh
r
a
,
“
A
n
a
i
v
e
a
p
p
r
o
a
c
h
:
t
r
a
n
s
l
a
t
i
o
n
o
f
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
t
o
st
r
u
c
t
u
r
e
d
q
u
e
r
y
l
a
n
g
u
a
g
e
,
”
i
n
I
C
S
T
T
r
a
n
sa
c
t
i
o
n
s
o
n
S
c
a
l
a
b
l
e
I
n
f
o
rm
a
t
i
o
n
S
y
s
t
e
m
s
,
2
0
2
3
,
d
o
i
:
1
0
.
4
1
0
8
/
e
e
t
s
i
s.
4
3
4
4
.
[
8
]
U
.
M
.
M
o
h
a
p
a
t
r
a
,
“
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
b
a
se
d
t
r
a
n
sl
a
t
i
o
n
f
r
o
m
O
D
I
A
t
o
H
I
N
D
I
,
”
J
o
u
rn
a
l
o
f
A
d
v
a
n
c
e
d
Mu
l
t
i
d
i
sc
i
p
l
i
n
a
ry
R
e
se
a
rc
h
S
t
u
d
i
e
s
a
n
d
D
e
v
e
l
o
p
m
e
n
t
,
v
o
l
.
4
,
n
o
.
2
,
p
p
.
1
1
8
–
1
2
5
,
2
0
2
2
5
.
[
9
]
P
.
M
.
S
u
b
h
a
s
h
,
C
.
R
.
K
a
v
i
t
h
a
,
D
.
G
u
p
t
a
,
a
n
d
V
.
K
a
n
j
i
r
a
n
g
a
t
,
“
I
n
d
o
-
A
r
y
a
n
d
i
a
l
e
c
t
i
d
e
n
t
i
f
i
c
a
t
i
o
n
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
e
n
se
mb
l
e
mo
d
e
l
,
”
Pro
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
2
3
5
,
n
o
.
C
,
p
p
.
2
8
8
6
–
28
9
6
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
2
4
.
0
4
.
2
7
3
.
[
1
0
]
A
.
N
a
i
k
,
M
.
K
a
r
a
n
i
,
K
.
C
h
h
e
d
a
,
a
n
d
M
.
G
a
d
a
,
“
N
M
T
f
o
r
E
n
g
l
i
s
h
-
M
a
r
a
t
h
i
u
si
n
g
R
N
N
s
&
a
t
t
e
n
t
i
o
n
,
”
i
n
2
0
2
2
6
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
on
C
o
m
p
u
t
i
n
g
,
C
o
m
m
u
n
i
c
a
t
i
o
n
,
C
o
n
t
ro
l
a
n
d
A
u
t
o
m
a
t
i
o
n
(
I
C
C
U
B
EA)
,
A
u
g
.
2
0
2
2
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
U
B
EA
5
4
9
9
2
.
2
0
2
2
.
1
0
0
1
0
7
8
2
.
[
1
1
]
S
.
Y
a
n
g
,
X
.
Y
u
,
a
n
d
Y
.
Z
h
o
u
,
“
LST
M
a
n
d
G
R
U
n
e
u
r
a
l
n
e
t
w
o
r
k
p
e
r
f
o
r
ma
n
c
e
c
o
mp
a
r
i
s
o
n
s
t
u
d
y
:
t
a
k
i
n
g
y
e
l
p
r
e
v
i
e
w
d
a
t
a
s
e
t
a
s
a
n
e
x
a
mp
l
e
,
”
i
n
2
0
2
0
I
n
t
e
r
n
a
t
i
o
n
a
l
W
o
rks
h
o
p
o
n
E
l
e
c
t
r
o
n
i
c
C
o
m
m
u
n
i
c
a
t
i
o
n
a
n
d
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
I
WE
C
AI
)
,
J
u
n
.
2
0
2
0
,
p
p
.
9
8
–
1
0
1
.
d
o
i
:
1
0
.
1
1
0
9
/
I
W
EC
A
I
5
0
9
5
6
.
2
0
2
0
.
0
0
0
2
7
.
[
1
2
]
D
.
D
a
t
t
a
,
P
.
E
.
D
a
v
i
d
,
D
.
M
i
t
t
a
l
,
a
n
d
A
.
Jai
n
,
“
N
e
u
r
a
l
ma
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
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
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
En
g
i
n
e
e
r
i
n
g
a
n
d
Ad
v
a
n
c
e
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
9
,
n
o
.
4
,
p
p
.
1
3
9
5
–
1
4
0
0
,
2
0
2
0
,
d
o
i
:
1
0
.
3
5
9
4
0
/
i
j
e
a
t
.
D
7
6
3
7
.
0
4
9
4
2
0
.
[
1
3
]
G
.
R
a
mes
h
e
t
a
l
.
,
“
S
a
m
a
n
a
n
t
a
r
:
t
h
e
l
a
r
g
e
st
p
u
b
l
i
c
l
y
a
v
a
i
l
a
b
l
e
p
a
r
a
l
l
e
l
c
o
r
p
o
r
a
c
o
l
l
e
c
t
i
o
n
f
o
r
1
1
I
n
d
i
c
l
a
n
g
u
a
g
e
s
,
”
T
ra
n
s
a
c
t
i
o
n
s
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
,
v
o
l
.
1
0
,
p
p
.
1
4
5
–
1
6
2
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
6
2
/
t
a
c
l
_
a
_
0
0
4
5
2
.
[
1
4
]
N
.
S
.
D
a
sh
a
n
d
B
.
B
.
C
h
a
u
d
h
u
r
i
,
“
W
h
y
d
o
w
e
n
e
e
d
t
o
d
e
v
e
l
o
p
c
o
r
p
o
r
a
i
n
I
n
d
i
a
n
l
a
n
g
u
a
g
e
s,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
W
o
r
k
i
n
g
C
o
n
f
e
re
e
n
c
e
o
n
S
h
a
ri
n
g
C
a
p
a
b
i
l
i
t
y
i
n
L
o
c
a
l
i
sa
t
i
o
n
a
n
d
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
C
ALL
A
-
2
0
0
1
)
,
2
0
0
1
,
p
p
.
1
-
18
.
[
1
5
]
K
.
P
a
p
i
n
e
n
i
,
S
.
R
o
u
k
o
s
,
T.
W
a
r
d
,
a
n
d
W
.
J.
Zh
u
,
“
B
L
EU
:
a
m
e
t
h
o
d
f
o
r
a
u
t
o
m
a
t
i
c
e
v
a
l
u
a
t
i
o
n
o
f
ma
c
h
i
n
e
t
r
a
n
sl
a
t
i
o
n
,
”
i
n
4
0
t
h
An
n
u
a
l
M
e
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
)
,
2
0
0
1
,
p
p
.
311
–
3
1
8
,
d
o
i
:
1
0
.
3
1
1
5
/
1
0
7
3
0
8
3
.
1
0
7
3
1
3
5
.
[
1
6
]
S
.
B
a
n
e
r
j
e
e
a
n
d
A
.
La
v
i
e
,
“
M
ETEO
R
:
a
n
a
u
t
o
m
a
t
i
c
m
e
t
r
i
c
f
o
r
M
T
e
v
a
l
u
a
t
i
o
n
w
i
t
h
i
mp
r
o
v
e
d
c
o
r
r
e
l
a
t
i
o
n
w
i
t
h
h
u
man
j
u
d
g
me
n
t
s,”
i
n
AC
L
W
o
rks
h
o
p
o
n
I
n
t
r
i
n
si
c
a
n
d
Ex
t
ri
n
si
c
E
v
a
l
u
a
t
i
o
n
M
e
a
su
r
e
s
f
o
r
M
a
c
h
i
n
e
T
ra
n
sl
a
t
i
o
n
a
n
d
/
o
r
S
u
m
m
a
ri
z
a
t
i
o
n
,
2
0
0
5
,
p
p
.
6
5
–
72
,
d
o
i
:
1
0
.
5
5
5
5
/
1
6
2
6
3
5
5
.
1
6
2
6
3
8
9
.
[
1
7
]
C
.
P
a
r
k
,
M
.
C
h
e
n
,
a
n
d
T
.
H
a
i
n
,
“
A
u
t
o
m
a
t
i
c
sp
e
e
c
h
r
e
c
o
g
n
i
t
i
o
n
s
y
st
e
m
-
i
n
d
e
p
e
n
d
e
n
t
w
o
r
d
e
r
r
o
r
r
a
t
e
e
st
i
ma
t
i
o
n
,
”
i
n
2
0
2
4
J
o
i
n
t
I
n
t
e
r
n
a
t
i
o
n
a
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
L
i
n
g
u
i
s
t
i
c
s,
L
a
n
g
u
a
g
e
Re
s
o
u
rc
e
s
a
n
d
Ev
a
l
u
a
t
i
o
n
(
L
RE
C
-
C
O
L
I
N
G
2
0
2
4
)
,
2
0
2
4
,
p
p
.
1
9
7
9
–
1
9
8
7
.
[
1
8
]
D
.
K
.
T
h
e
n
n
a
l
,
J.
Jam
e
s,
D
.
P
.
G
o
p
i
n
a
t
h
,
a
n
d
M
.
K
.
A
sh
r
a
f
,
“
A
d
v
o
c
a
t
i
n
g
c
h
a
r
a
c
t
e
r
e
r
r
o
r
r
a
t
e
f
o
r
m
u
l
t
i
l
i
n
g
u
a
l
A
S
R
e
v
a
l
u
a
t
i
o
n
,
”
i
n
An
n
u
a
l
C
o
n
f
e
r
e
n
c
e
o
f
t
h
e
N
a
t
i
o
n
s
o
f
t
h
e
Am
e
ri
c
a
s
C
h
a
p
t
e
r
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
st
i
c
s
,
2
0
2
5
,
p
p
.
4
9
2
6
–
4
9
3
5
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
5
.
f
i
n
d
i
n
g
s
-
n
a
a
c
l
.
2
7
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
A
s
s
es
s
men
t o
f d
ee
p
lea
r
n
in
g
b
a
s
ed
Hin
d
i O
d
ia
b
id
ir
ec
tio
n
a
l
ma
ch
in
e
tr
a
n
s
la
tio
n
…
(
S
u
b
h
a
s
h
r
ee
S
a
tp
a
th
y)
695
[
1
9
]
R
.
R
a
j
a
a
n
d
A
.
V
a
t
s,
“
P
a
r
a
l
l
e
l
c
o
r
p
o
r
a
f
o
r
ma
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
i
n
l
o
w
-
r
e
so
u
r
c
e
I
n
d
i
c
l
a
n
g
u
a
g
e
s:
a
c
o
m
p
r
e
h
e
n
si
v
e
r
e
v
i
e
w
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
E
i
g
h
t
h
W
o
r
k
sh
o
p
o
n
T
e
c
h
n
o
l
o
g
i
e
s
f
o
r
Ma
c
h
i
n
e
T
ra
n
s
l
a
t
i
o
n
o
f
L
o
w
-
Re
s
o
u
rc
e
L
a
n
g
u
a
g
e
s
(
L
o
Re
s
MT
2
0
2
5
)
,
S
t
r
o
u
d
s
b
u
r
g
,
U
S
A
,
2
0
2
5
,
p
p
.
1
2
9
–
1
4
3
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
5
.
l
o
r
e
smt
-
1
.
1
2
.
[
2
0
]
S
.
B
.
D
a
s
,
A
.
B
i
r
a
d
a
r
,
T.
K
.
M
i
sh
r
a
,
a
n
d
B
.
K
.
P
a
t
r
a
,
“
I
mp
r
o
v
i
n
g
m
u
l
t
i
l
i
n
g
u
a
l
n
e
u
r
a
l
m
a
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
sy
st
e
m
f
o
r
I
n
d
i
c
l
a
n
g
u
a
g
e
s
,
”
A
C
M
T
r
a
n
s
a
c
t
i
o
n
s
o
n
Asi
a
n
a
n
d
L
o
w
-
Re
s
o
u
rce
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
2
,
n
o
.
6
,
p
p
.
1
–
2
4
,
Ju
n
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
4
5
/
3
5
8
7
9
3
2
.
[
2
1
]
B
.
H
a
d
d
o
w
,
R
.
B
a
w
d
e
n
,
A
.
V
.
M
.
B
a
r
o
n
e
,
J.
H
e
l
c
l
,
a
n
d
A
.
B
i
r
c
h
,
“
S
u
r
v
e
y
o
f
l
o
w
-
r
e
s
o
u
r
c
e
ma
c
h
i
n
e
t
r
a
n
sl
a
t
i
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
,
v
o
l
.
4
8
,
n
o
.
3
,
p
p
.
6
7
3
–
7
3
2
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
6
2
/
c
o
l
i
_
a
_
0
0
4
4
6
.
[
2
2
]
A
.
G
o
d
a
s
e
a
n
d
S
.
G
o
v
i
l
k
a
r
,
“
M
a
c
h
i
n
e
t
r
a
n
sl
a
t
i
o
n
d
e
v
e
l
o
p
m
e
n
t
f
o
r
I
n
d
i
a
n
l
a
n
g
u
a
g
e
s
a
n
d
i
t
s
a
p
p
r
o
a
c
h
e
s,”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
C
o
m
p
u
t
i
n
g
,
v
o
l
.
4
,
n
o
.
2
,
p
p
.
5
5
–
7
4
,
2
0
1
5
,
d
o
i
:
1
0
.
5
1
2
1
/
i
j
n
l
c
.
2
0
1
5
.
4
2
0
5
.
[
2
3
]
Ta
t
o
e
b
a
P
r
o
j
e
c
t
,
“
T
a
b
-
d
e
l
i
m
i
t
e
d
b
i
l
i
n
g
u
a
l
se
n
t
e
n
c
e
p
a
i
r
s,”
M
a
n
y
Th
i
n
g
s
.
o
r
g
.
A
c
c
e
ss
e
d
:
Ja
n
.
2
0
,
2
0
2
5
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
ma
n
y
t
h
i
n
g
s.
o
r
g
/
a
n
k
i
/
[
2
4
]
K
.
C
h
o
e
t
a
l
.
,
“
L
e
a
r
n
i
n
g
p
h
r
a
s
e
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
u
si
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
s
t
i
c
a
l
mac
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
,
”
i
n
EM
N
L
P
2
0
1
4
-
2
0
1
4
C
o
n
f
e
r
e
n
c
e
o
n
Em
p
i
ri
c
a
l
M
e
t
h
o
d
s
i
n
N
a
t
u
ra
l
L
a
n
g
u
a
g
e
Pr
o
c
e
ssi
n
g
,
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
C
o
n
f
e
re
n
c
e
,
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
.
[
2
5
]
Y
.
B
e
n
g
i
o
,
P
.
S
i
mar
d
,
a
n
d
P
.
F
r
a
sc
o
n
i
,
“
L
e
a
r
n
i
n
g
l
o
n
g
-
t
e
r
m
d
e
p
e
n
d
e
n
c
i
e
s
w
i
t
h
g
r
a
d
i
e
n
t
d
e
sce
n
t
i
s
d
i
f
f
i
c
u
l
t
,
”
I
E
EE
T
r
a
n
sa
c
t
i
o
n
s
o
n
N
e
u
r
a
l
N
e
t
w
o
r
k
s
,
v
o
l
.
5
,
n
o
.
2
,
p
p
.
1
5
7
–
1
6
6
,
M
a
r
.
1
9
9
4
,
d
o
i
:
1
0
.
1
1
0
9
/
7
2
.
2
7
9
1
8
1
.
[
2
6
]
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
,
“
Lo
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
.
[
2
7
]
M
.
S
u
n
d
e
r
me
y
e
r
,
R
.
S
c
h
l
ü
t
e
r
,
a
n
d
H
.
N
e
y
,
“
LS
TM
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
l
a
n
g
u
a
g
e
mo
d
e
l
i
n
g
,
”
i
n
I
n
t
e
rs
p
e
e
c
h
2
0
1
2
,
S
e
p
.
2
0
1
2
,
p
p
.
1
9
4
–
1
9
7
.
d
o
i
:
1
0
.
2
1
4
3
7
/
I
n
t
e
r
sp
e
e
c
h
.
2
0
1
2
-
6
5
.
[
2
8
]
J.
C
h
u
n
g
,
C
.
G
u
l
c
e
h
r
e
,
K
.
C
h
o
,
a
n
d
Y
.
B
e
n
g
i
o
,
“
Em
p
i
r
i
c
a
l
e
v
a
l
u
a
t
i
o
n
o
f
g
a
t
e
d
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
s
o
n
s
e
q
u
e
n
c
e
m
o
d
e
l
i
n
g
,
”
2
0
1
4
,
a
rX
i
v
:
1
4
1
2
.
3
5
5
5
.
[
2
9
]
I
.
S
u
t
sk
e
v
e
r
,
O
.
V
i
n
y
a
l
s
,
a
n
d
Q
.
V
.
L
e
,
“
S
e
q
u
e
n
c
e
t
o
se
q
u
e
n
c
e
l
e
a
r
n
i
n
g
w
i
t
h
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
i
n
2
8
t
h
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
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ss
i
n
g
S
y
s
t
e
m
s
,
2
0
1
4
,
p
p
.
3
1
0
4
–
3
1
1
2
,
d
o
i
:
1
0
.
5
5
5
5
/
2
9
6
9
0
3
3
.
2
9
6
9
1
7
3
.
[
3
0
]
R
.
P
a
sc
a
n
u
,
C
.
G
u
l
c
e
h
r
e
,
K
.
C
h
o
,
a
n
d
Y
.
B
e
n
g
i
o
,
“
H
o
w
t
o
c
o
n
st
r
u
c
t
d
e
e
p
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
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
L
e
a
rn
i
n
g
Re
p
res
e
n
t
a
t
i
o
n
s,
B
a
n
f
f
,
C
a
n
a
d
a
,
2
0
1
4
.
[
3
1
]
K
.
R
e
v
a
n
u
r
u
,
K
.
T
u
r
l
a
p
a
t
y
,
a
n
d
S
.
R
a
o
,
“
N
e
u
r
a
l
mac
h
i
n
e
t
r
a
n
sl
a
t
i
o
n
o
f
I
n
d
i
a
n
l
a
n
g
u
a
g
e
s,
”
i
n
1
0
t
h
A
n
n
u
a
l
A
C
M
I
n
d
i
a
C
o
m
p
u
t
e
C
o
n
f
e
re
n
c
e
,
N
e
w
Y
o
r
k
,
U
n
i
t
e
d
S
t
a
t
e
s
,
N
o
v
.
2
0
1
7
,
p
p
.
1
1
–
20
,
d
o
i
:
1
0
.
1
1
4
5
/
3
1
4
0
1
0
7
.
3
1
4
0
1
1
1
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
S
u
b
h
a
shre
e
S
a
tp
a
t
h
y
re
c
e
iv
e
d
th
e
d
e
g
re
e
in
M
CA
fro
m
CM
R
Co
ll
e
g
e
o
f
En
g
i
n
e
e
rin
g
a
n
d
Tec
h
n
o
l
o
g
y
,
Hy
d
e
ra
b
a
d
(CM
RCET
/C
M
RK),
In
d
ia
i
n
2
0
1
6
.
M
.
Tec
h
.
in
Co
m
p
u
ter
S
c
ien
c
e
fr
o
m
De
p
a
rtme
n
t
o
f
C
o
m
p
u
ter
S
c
ien
c
e
a
n
d
I
n
fo
rm
a
ti
o
n
Tec
h
n
o
lo
g
y
,
S
i
k
sh
a
‘O’
An
u
sa
n
d
h
a
n
(De
e
m
e
d
to
b
e
Un
iv
e
rsit
y
)
,
Bh
u
b
a
n
e
sw
a
r,
In
d
i
a
in
2
0
1
8
.
S
h
e
is
c
u
rre
n
tl
y
p
u
rsu
i
n
g
h
is
P
h
.
D.
i
n
C
o
m
p
u
ter
S
c
ien
c
e
a
n
d
E
n
g
in
e
e
rin
g
a
t
S
ik
s
h
a
‘O’
An
u
sa
n
d
h
a
n
(De
e
m
e
d
to
b
e
U
n
iv
e
rsit
y
)
,
B
h
u
b
a
n
e
sw
a
r,
In
d
ia.
S
h
e
h
a
s
p
u
b
li
s
h
e
d
5
p
a
p
e
rs
in
v
a
ri
o
u
s
i
n
tern
a
ti
o
n
a
l
c
o
n
fe
re
n
c
e
s.
S
h
e
c
a
n
b
e
c
o
n
tac
te
d
a
t
e
m
a
il
:
su
b
h
a
sh
re
e
sa
tp
a
th
y
1
6
@g
m
a
il
.
c
o
m
.
Pro
f.
Dr
.
S
m
it
a
p
r
a
v
a
Mi
shr
a
is
c
u
rre
n
tl
y
wo
r
k
in
g
a
s
p
ro
fe
ss
o
r
in
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
In
f
o
rm
a
ti
o
n
Tec
h
n
o
lo
g
y
,
S
i
k
sh
a
‘O’
An
u
sa
n
d
h
a
n
(De
e
m
e
d
to
b
e
Un
iv
e
rsity
)
,
Bh
u
b
a
n
e
sw
a
r,
In
d
ia
.
S
h
e
h
a
s
1
8
y
e
a
rs
o
f
tea
c
h
i
n
g
a
n
d
re
se
a
rc
h
e
x
p
e
rien
c
e
in
t
h
e
c
u
rre
n
t
o
rg
a
n
iza
ti
o
n
.
S
h
e
h
a
s
p
u
b
li
s
h
e
d
3
0
n
u
m
b
e
rs
o
f
re
se
a
rc
h
p
a
p
e
rs
in
v
a
ri
o
u
s
re
p
u
ted
jo
u
r
n
a
ls
a
n
d
c
o
n
fe
re
n
c
e
s.
1
5
M
.
Tec
h
.
s
c
h
o
lars
p
r
o
d
u
c
e
d
u
n
d
e
r
h
e
r
su
p
e
rv
isi
o
n
i
n
t
h
e
r
e
se
a
rc
h
a
re
a
s
o
f
d
a
ta
m
in
i
n
g
,
m
a
c
h
in
e
le
a
rn
in
g
,
a
n
d
n
a
t
u
ra
l
la
n
g
u
a
g
e
p
ro
c
e
ss
in
g
.
4
P
h
.
D.
sc
h
o
lars
a
re
p
u
rsu
i
n
g
re
se
a
rc
h
u
n
d
e
r
h
e
r
s
u
p
e
rv
isio
n
.
S
h
e
h
a
s
c
o
n
tri
b
u
ted
se
v
e
ra
l
a
c
a
d
e
m
ic
a
c
ti
v
it
ies
a
t
o
rg
a
n
iza
ti
o
n
lev
e
l.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
sm
it
a
m
ish
ra
@s
o
a
.
a
c
.
in
.
Pro
f.
Dr
.
Ajit
K
u
m
a
r
Na
y
a
k
is
th
e
p
ro
fe
ss
o
r
a
n
d
h
e
a
d
o
f
t
h
e
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
In
f
o
rm
a
ti
o
n
Tec
h
n
o
lo
g
y
,
S
i
k
sh
a
‘O’
An
u
sa
n
d
h
a
n
(De
e
m
e
d
to
b
e
Un
iv
e
rsity
)
,
B
h
u
b
a
n
e
sw
a
r,
In
d
i
a
.
He
re
c
e
iv
e
d
d
e
g
re
e
El
e
c
tri
c
a
l
in
En
g
in
e
e
rin
g
fr
o
m
t
h
e
In
stit
u
t
io
n
o
f
E
n
g
in
e
e
rs,
In
d
ia
i
n
th
e
y
e
a
r
1
9
9
4
,
M
.
Tec
h
.
a
n
d
P
h
.
D.
d
e
g
re
e
s
in
Co
m
p
u
ter
S
c
ien
c
e
fro
m
Utk
a
l
Un
i
v
e
rsity
in
2
0
0
1
a
n
d
2
0
1
0
re
sp
e
c
ti
v
e
l
y
.
He
h
a
s
p
u
b
li
s
h
e
d
a
b
o
u
t
7
0
re
se
a
rc
h
p
a
p
e
rs
in
v
a
rio
u
s
j
o
u
rn
a
ls
a
n
d
c
o
n
fe
re
n
c
e
s.
Also
c
o
-
a
u
th
o
re
d
a
b
o
o
k
‘
C
o
m
p
u
ter
n
e
two
rk
sim
u
lati
o
n
u
sin
g
NS2
’
.
10
P
h
.
D.
sc
h
o
lars
h
a
v
e
b
e
e
n
a
wa
rd
e
d
P
h
.
D.
u
n
d
e
r
h
i
s
su
p
e
rv
isi
o
n
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
ji
tn
a
y
a
k
@s
o
a
.
a
c
.
i
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.