I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
1
6
,
No
.
2
,
A
p
r
il
20
2
6
,
p
p
.
973
~
980
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
1
6
i
2
.
pp
9
7
3
-
9
8
0
973
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Cro
ss
-
ling
ua
l sem
a
ntic
a
lig
nment
and trans
f
er
lea
rni
ng
using
multiling
ua
l lang
ua
g
e mo
dels
Nira
nja
n G
C,
Ra
ma
ka
nth
K
um
a
r
P
,
P
a
v
it
hra
H
,
M
ina
l Mo
ha
rir
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
r
i
n
g
,
R
V
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
,
B
e
n
g
a
l
u
r
u
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
1
,
2
0
2
5
R
ev
is
ed
Dec
2
1
,
2
0
2
5
Acc
ep
ted
J
an
1
6
,
2
0
2
6
M
u
lt
il
in
g
u
a
l
lan
g
u
a
g
e
m
o
d
e
ls
(
M
LM
s)
a
re
wid
e
ly
u
se
d
fo
r
c
r
o
ss
-
li
n
g
u
a
l
tas
k
s,
y
e
t
th
e
ir
a
b
il
it
y
t
o
a
c
h
ie
v
e
c
o
n
siste
n
t
se
m
a
n
ti
c
a
li
g
n
m
e
n
t
a
n
d
tran
sfe
r
to
l
o
w
-
re
so
u
rc
e
lan
g
u
a
g
e
s
re
m
a
in
s
li
m
it
e
d
.
T
h
is
wo
r
k
e
x
a
m
in
e
s
c
ro
ss
-
li
n
g
u
a
l
se
m
a
n
ti
c
a
li
g
n
m
e
n
t
a
n
d
tran
sfe
r
lea
rn
in
g
th
r
o
u
g
h
a
c
o
m
p
a
ra
ti
v
e
e
v
a
lu
a
ti
o
n
o
f
M
LM
s
a
t
b
o
th
th
e
wo
rd
a
n
d
se
n
ten
c
e
lev
e
ls.
W
e
a
n
a
ly
z
e
g
e
n
e
ra
l
-
p
u
r
p
o
se
m
o
d
e
ls
su
c
h
a
s
b
ig
sc
ien
c
e
larg
e
o
p
e
n
-
sc
ien
c
e
o
p
e
n
-
a
c
c
e
ss
m
u
lt
il
in
g
u
a
l
la
n
g
u
a
g
e
m
o
d
e
l
(
BLOOM
)
a
n
d
tas
k
-
sp
e
c
ialize
d
m
o
d
e
ls
in
c
lu
d
in
g
LaBSE
a
n
d
XLM
-
R
a
c
ro
ss
En
g
li
sh
,
F
re
n
c
h
,
Hin
d
i,
a
n
d
Ka
n
n
a
d
a
.
Wo
rd
-
le
v
e
l
e
x
p
e
rime
n
ts
sh
o
w
t
h
a
t
LaBSE
a
c
h
ie
v
e
s
su
b
sta
n
ti
a
l
ly
h
ig
h
e
r
c
o
sin
e
sim
il
a
rit
y
sc
o
re
s
o
f
a
b
o
v
e
0
.
8
0
a
c
ro
ss
lan
g
u
a
g
e
s.
I
n
se
n
t
e
n
c
e
-
lev
e
l
n
a
tu
ra
l
la
n
g
u
a
g
e
i
n
fe
re
n
c
e
,
XLM
-
R
o
u
tp
e
rfo
rm
s
o
t
h
e
r
m
o
d
e
ls,
a
c
h
iev
i
n
g
a
n
F
1
sc
o
re
o
f
6
8
.
6
2
%
o
n
Ka
n
n
a
d
a
a
n
d
7
4
.
8
1
%
o
n
F
re
n
c
h
.
Th
e
se
re
su
lt
s
in
d
ica
te
th
a
t
m
o
d
e
l
s
p
e
c
ializa
ti
o
n
a
n
d
trai
n
in
g
o
b
jec
ti
v
e
s
p
lay
a
c
r
u
c
ial
ro
le
in
c
ro
ss
-
li
n
g
u
a
l
p
e
rf
o
rm
a
n
c
e
,
p
a
rti
c
u
larly
fo
r
lo
w
-
re
so
u
rc
e
la
n
g
u
a
g
e
s,
a
n
d
sh
o
u
l
d
b
e
c
a
re
fu
ll
y
c
o
n
sid
e
re
d
wh
e
n
d
e
p
lo
y
in
g
m
u
lt
il
i
n
g
u
a
l
n
a
tu
ra
l
lan
g
u
a
g
e
p
r
o
c
e
ss
in
g
(NLP
)
sy
ste
m
s.
K
ey
w
o
r
d
s
:
C
o
s
in
e
s
im
ilar
ity
Mu
ltil
in
g
u
al
lan
g
u
a
g
e
m
o
d
els
Natu
r
al
lan
g
u
ag
e
in
f
er
e
n
ce
Natu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
Sem
an
tic
alig
n
m
en
t
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
:
Nir
an
jan
G
C
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
,
R
V
C
o
lleg
e
o
f
E
n
g
in
ee
r
i
n
g
My
s
o
r
e
R
o
ad
,
R
V
Vid
y
an
ik
et
an
Po
s
t,
B
en
g
alu
r
u
,
Kar
n
atak
a
5
6
0
0
5
9
,
I
n
d
ia
E
m
ail:
n
ir
an
jan
g
c.
s
cs2
3
@
r
v
ce
.
ed
u
.
in
1.
I
NT
RO
D
UCT
I
O
N
Ar
tific
ial
i
n
tellig
en
ce
(
AI
)
h
as
b
r
o
u
g
h
t
ab
o
u
t
s
ig
n
if
ican
t
ch
an
g
es
in
th
e
way
we
in
ter
ac
t
with
tech
n
o
lo
g
y
to
d
ay
.
Am
o
n
g
th
e
v
ar
io
u
s
b
r
an
c
h
es o
f
AI
,
n
atu
r
a
l la
n
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P)
h
as st
o
o
d
o
u
t f
o
r
its
ab
ilit
y
to
h
elp
m
ac
h
in
es
u
n
d
e
r
s
tan
d
,
in
ter
p
r
et,
an
d
g
en
e
r
ate
h
u
m
an
lan
g
u
ag
e.
W
ith
th
e
r
is
e
o
f
g
en
e
r
ativ
e
AI
,
th
e
ca
p
ab
ilit
ies
o
f
lan
g
u
ag
e
m
o
d
els
h
av
e
g
r
o
wn
r
ap
id
ly
,
en
a
b
lin
g
m
ac
h
in
es
to
p
er
f
o
r
m
task
s
lik
e
tr
an
s
latio
n
,
s
u
m
m
ar
izatio
n
,
an
d
ev
en
cr
ea
tiv
e
wr
itin
g
.
On
e
o
f
th
e
m
o
r
e
r
ec
en
t
a
d
v
an
ce
m
en
ts
in
th
is
s
p
ac
e
is
th
e
d
ev
elo
p
m
e
n
t
o
f
m
u
ltil
in
g
u
al
lan
g
u
ag
e
m
o
d
els
(
ML
Ms)
-
m
o
d
els
th
at
ca
n
p
r
o
ce
s
s
an
d
u
n
d
er
s
tan
d
m
u
ltip
le
lan
g
u
ag
es u
s
in
g
a
s
h
ar
ed
n
eu
r
al
ar
ch
itectu
r
e.
I
n
m
u
ltil
in
g
u
al
s
ettin
g
s
,
th
ese
m
o
d
els
ar
e
esp
ec
ially
u
s
ef
u
l
b
ec
au
s
e
th
ey
ca
n
h
an
d
le
i
n
p
u
t
f
r
o
m
d
if
f
er
en
t
lan
g
u
ag
es
with
o
u
t
r
e
q
u
ir
in
g
s
ep
ar
ate
m
o
d
els
f
o
r
e
ac
h
o
n
e.
T
h
is
m
ak
es
th
em
id
e
al
f
o
r
u
s
e
i
n
g
lo
b
al
ap
p
licatio
n
s
wh
er
e
lan
g
u
ag
e
d
iv
er
s
ity
is
co
m
m
o
n
.
I
n
th
e
c
o
n
tex
t
o
f
t
h
is
p
r
o
ject,
we
ex
p
lo
r
e
h
o
w
ML
Ms
ca
n
b
e
ap
p
lied
to
u
n
d
er
s
tan
d
an
d
p
r
o
ce
s
s
d
if
f
er
en
t
lan
g
u
ag
es
ef
f
ec
tiv
ely
.
T
h
e
p
r
o
ject
aim
s
to
an
aly
ze
wh
eth
er
th
ese
m
o
d
els
tr
u
ly
ca
p
tu
r
e
th
e
m
ea
n
in
g
o
f
wo
r
d
s
ac
r
o
s
s
lan
g
u
ag
es
an
d
h
o
w
well
th
ey
ca
n
tr
an
s
f
er
k
n
o
wled
g
e
f
r
o
m
a
h
ig
h
r
eso
u
r
ce
lan
g
u
ag
e
(
E
n
g
lis
h
)
to
a
lo
w
r
eso
u
r
ce
I
n
d
ic
l
an
g
u
a
g
e
(
Kan
n
a
d
a)
.
MLMs
s
u
ch
as
B
L
O
OM
-
1
.
7
B
an
d
QW
E
N2
h
av
e
b
ee
n
ev
alu
ated
f
o
r
th
eir
ab
ilit
y
to
alig
n
cr
o
s
s
-
lin
g
u
al
wo
r
d
em
b
ed
d
in
g
s
,
r
e
p
r
esen
t
in
ter
n
al
s
tr
u
ctu
r
es
th
r
o
u
g
h
p
r
o
b
in
g
task
s
lik
e
s
en
t
en
ce
s
im
ilar
ity
an
d
n
am
ed
en
tity
r
ec
o
g
n
itio
n
,
a
n
d
tr
an
s
f
er
k
n
o
wled
g
e
ac
r
o
s
s
lan
g
u
ag
es,
with
a
f
o
cu
s
o
n
lo
w
-
r
eso
u
r
ce
s
ce
n
ar
io
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
2
,
Ap
r
il
20
2
6
:
9
7
3
-
980
974
[
1
]
.
E
v
alu
atio
n
m
eth
o
d
o
lo
g
i
es
f
o
r
m
o
d
els
lik
e
m
B
E
R
T
,
XL
M
-
R
,
an
d
I
n
f
o
XL
M
h
av
e
r
ev
ea
led
th
at
p
er
f
o
r
m
an
ce
s
ig
n
if
ican
tly
d
r
o
p
s
in
ac
r
o
s
s
-
lan
g
u
ag
e
task
s
co
m
p
ar
ed
t
o
with
in
-
lan
g
u
ag
e
s
etu
p
s
,
esp
ec
ially
in
lo
w
-
r
eso
u
r
ce
la
n
g
u
a
g
es,
ex
p
o
s
in
g
g
en
er
aliza
tio
n
wea
k
n
ess
es
[
2
]
.
T
h
e
t
r
an
s
latio
n
p
e
r
f
o
r
m
an
ce
o
f
b
ig
s
cien
ce
lar
g
e
o
p
en
-
s
cien
ce
o
p
en
-
ac
ce
s
s
m
u
ltil
in
g
u
al
lan
g
u
ag
e
m
o
d
el
(
B
L
OOM
)
h
as
b
ee
n
as
s
es
s
ed
in
ze
r
o
-
s
h
o
t
an
d
f
ew
-
s
h
o
t
s
ettin
g
s
ac
r
o
s
s
h
ig
h
-
an
d
l
o
w
-
r
eso
u
r
ce
lan
g
u
ag
e
p
air
s
,
s
h
o
win
g
n
o
tab
le
im
p
r
o
v
em
en
ts
in
f
ew
-
s
h
o
t
s
ce
n
ar
io
s
wh
en
lin
g
u
is
tic
s
im
ilar
ities
ex
is
t
b
etwe
en
s
o
u
r
ce
an
d
tar
g
et
lan
g
u
ag
es
[
3
]
.
A
m
o
r
e
ef
f
icien
t
ap
p
r
o
ac
h
to
cr
o
s
s
-
lin
g
u
al
p
r
etr
ain
in
g
h
as
b
ee
n
in
tr
o
d
u
ce
d
with
XL
M
-
E
[
4
]
,
w
h
ich
u
s
es
r
ep
lace
d
to
k
e
n
d
etec
tio
n
in
s
tead
o
f
m
ask
e
d
la
n
g
u
ag
e
m
o
d
elin
g
a
n
d
ac
h
iev
e
s
f
aster
,
co
s
t
-
ef
f
ec
tiv
e
tr
ai
n
in
g
wh
ile
m
ain
tain
in
g
h
ig
h
p
e
r
f
o
r
m
an
ce
o
n
b
e
n
ch
m
a
r
k
s
lik
e
XT
R
E
ME
[
5
]
.
Mu
ltil
in
g
u
al
B
E
R
T
h
as
s
h
o
wn
s
tr
o
n
g
p
er
f
o
r
m
an
ce
in
h
ig
h
-
r
eso
u
r
ce
lan
g
u
a
g
es
f
o
r
n
a
m
ed
en
tity
r
ec
o
g
n
itio
n
,
b
u
t
its
ef
f
ec
tiv
e
n
ess
s
h
ar
p
ly
d
ec
lin
es
i
n
lo
w
-
r
eso
u
r
ce
la
n
g
u
a
g
es
d
u
e
to
l
im
ited
v
o
ca
b
u
lar
y
co
v
er
ag
e
a
n
d
in
s
u
f
f
icien
t
tr
ai
n
in
g
d
ata,
wh
ich
ar
e
i
d
en
tifie
d
as
cr
itical
f
ac
to
r
s
af
f
ec
tin
g
m
u
ltil
in
g
u
al
m
o
d
el
p
er
f
o
r
m
an
ce
[
6
]
.
Neg
ativ
e
in
ter
f
er
en
ce
,
wh
er
e
m
u
ltil
in
g
u
a
l
tr
ain
in
g
d
eg
r
ad
es
p
e
r
f
o
r
m
a
n
ce
o
n
in
d
iv
id
u
al
lan
g
u
ag
es,
h
as
b
ee
n
a
d
d
r
ess
ed
th
r
o
u
g
h
a
m
eta
-
lear
n
in
g
ap
p
r
o
ac
h
th
at
in
tr
o
d
u
ce
s
lan
g
u
a
g
e
-
s
p
ec
if
ic
ad
ap
ter
s
to
im
p
r
o
v
e
b
o
t
h
m
o
n
o
lin
g
u
al
ac
cu
r
ac
y
a
n
d
cr
o
s
s
-
lin
g
u
al
t
r
an
s
f
er
[
7
]
.
A
co
m
p
lem
en
tar
y
s
tr
ateg
y
in
v
o
lv
es
u
s
in
g
lan
g
u
a
g
e
-
s
p
ec
if
ic
s
u
b
n
e
two
r
k
s
with
in
m
u
ltil
in
g
u
al
m
o
d
els
to
co
n
tr
o
l
p
ar
a
m
eter
s
h
ar
in
g
,
wh
ich
,
wh
en
co
m
b
in
ed
with
m
eta
-
lear
n
i
n
g
,
y
ield
s
s
ig
n
if
ican
t
im
p
r
o
v
em
en
ts
in
f
ew
-
s
h
o
t
tr
a
n
s
f
er
an
d
lo
w
-
r
eso
u
r
ce
d
ep
en
d
e
n
cy
p
ar
s
in
g
[
8
]
.
I
n
th
e
d
o
m
ain
o
f
cr
o
s
s
-
lin
g
u
al
in
f
o
r
m
atio
n
r
etr
iev
al,
a
n
o
v
el
m
eth
o
d
ca
lled
OPTI
C
AL
u
s
es
o
p
tim
al
tr
an
s
p
o
r
t
-
b
ased
k
n
o
wled
g
e
d
is
till
atio
n
to
alig
n
to
k
en
-
lev
el
e
m
b
ed
d
in
g
s
b
etwe
en
lan
g
u
ag
es,
e
n
ab
lin
g
h
ig
h
r
etr
iev
al
p
er
f
o
r
m
an
ce
in
lo
w
-
r
es
o
u
r
ce
s
ettin
g
s
with
o
u
t
r
ely
in
g
o
n
cr
o
s
s
-
lin
g
u
al
r
elev
an
ce
an
n
o
tatio
n
s
[
9
]
.
T
h
e
r
o
b
u
s
tn
ess
o
f
m
u
ltil
in
g
u
a
l
m
o
d
els
lik
e
m
B
E
R
T
an
d
XL
M
-
R
u
n
d
er
ad
v
e
r
s
ar
ial
p
er
tu
r
b
atio
n
s
h
as
b
ee
n
ev
al
u
ated
f
o
r
task
s
s
u
ch
as
n
am
ed
e
n
tity
r
ec
o
g
n
itio
n
,
r
ev
ea
lin
g
th
at
v
o
ca
b
u
lar
y
o
v
er
lap
an
d
lin
g
u
is
tic
p
r
o
x
im
ity
a
r
e
k
ey
d
eter
m
in
a
n
ts
o
f
cr
o
s
s
-
lin
g
u
al
g
e
n
er
aliza
tio
n
in
lo
w
-
r
eso
u
r
ce
s
ettin
g
s
[
1
0
]
.
A
co
m
p
ar
ativ
e
s
tu
d
y
in
v
o
lv
i
n
g
m
u
ltil
in
g
u
al
an
d
m
o
n
o
lin
g
u
al
m
o
d
els
o
n
Af
r
ican
lan
g
u
ag
es
Kin
y
ar
wan
d
a
an
d
Kir
u
n
d
i
s
h
o
wed
th
at
f
in
e
-
tu
n
e
d
m
u
ltil
in
g
u
al
m
o
d
els
lik
e
Af
r
iB
E
R
T
o
u
tp
er
f
o
r
m
o
t
h
er
s
,
d
em
o
n
s
tr
atin
g
s
tr
o
n
g
tr
an
s
f
e
r
ca
p
ab
ilit
ies
b
etwe
en
lin
g
u
is
tically
s
im
ilar
,
lo
w
-
r
eso
u
r
ce
lan
g
u
ag
es
[
1
1
]
.
A
s
tr
u
ctu
r
e
d
s
u
r
v
ey
o
f
c
r
o
s
s
-
lin
g
u
al
wo
r
d
em
b
ed
d
in
g
tech
n
iq
u
es c
ateg
o
r
ized
m
eth
o
d
s
b
y
d
ata
ali
g
n
m
en
t le
v
els an
d
ty
p
e
(
p
ar
all
el
v
s
.
co
m
p
ar
ab
le)
,
o
f
f
er
in
g
a
f
o
u
n
d
atio
n
al
u
n
d
e
r
s
tan
d
in
g
o
f
m
u
ltil
in
g
u
al
s
em
an
tic
r
ep
r
esen
tatio
n
ap
p
r
o
a
ch
es
[
1
2
]
.
T
r
a
n
s
f
er
lear
n
in
g
s
tr
ateg
ies
ac
r
o
s
s
v
a
r
io
u
s
m
u
ltil
in
g
u
al
NL
P
task
s
h
av
e
b
ee
n
r
ev
iewe
d
,
h
ig
h
l
ig
h
tin
g
th
at
m
o
d
e
l
ar
ch
itectu
r
e
an
d
p
r
et
r
ain
in
g
d
ata
ar
e
m
o
r
e
im
p
ac
tf
u
l
th
an
s
cr
ip
t
s
im
ilar
ity
,
w
ith
m
o
d
els
lik
e
XL
M
-
R
o
u
t
p
er
f
o
r
m
in
g
b
o
th
m
B
E
R
T
an
d
lan
g
u
ag
e
-
s
p
ec
if
ic
v
ar
ian
ts
in
z
er
o
-
s
h
o
t tr
a
n
s
f
er
s
ce
n
ar
io
s
[
1
3
]
.
Z
er
o
Sh
o
tTM
d
em
o
n
s
tr
ated
th
at
m
u
ltil
in
g
u
al
co
n
tex
tu
al
ized
em
b
ed
d
in
g
s
ca
n
b
e
ef
f
ec
tiv
ely
lev
er
ag
ed
f
o
r
ze
r
o
-
s
h
o
t
t
o
p
ic
m
o
d
elin
g
,
en
a
b
lin
g
c
r
o
s
s
-
lin
g
u
al
to
p
ic
in
f
er
e
n
ce
with
o
u
t
r
et
r
ain
in
g
o
r
lan
g
u
a
g
e
-
s
p
ec
if
ic
v
o
ca
b
u
lar
ies
[
1
4
]
.
I
n
th
e
d
o
m
ain
o
f
lo
w
-
r
eso
u
r
c
e
I
n
d
ia
n
lan
g
u
ag
es,
ze
r
o
-
s
h
o
t
tr
an
s
latio
n
u
s
in
g
m
u
ltil
in
g
u
al
n
e
u
r
al
m
ac
h
in
e
tr
an
s
latio
n
(
NM
T
)
m
o
d
els
s
h
o
wed
s
u
b
s
tan
tial
im
p
r
o
v
em
en
t
wh
en
in
co
r
p
o
r
atin
g
tr
ain
in
g
d
ata
f
r
o
m
r
elate
d
la
n
g
u
ag
es,
e
m
p
h
asizin
g
th
e
r
o
l
e
o
f
le
x
ical
p
r
o
x
im
ity
in
en
h
an
cin
g
tr
an
s
latio
n
p
er
f
o
r
m
an
ce
[
1
5
]
.
T
o
s
u
p
p
o
r
t
NL
P
d
ev
elo
p
m
e
n
t
f
o
r
I
n
d
ia
n
lan
g
u
ag
es,
th
e
I
n
d
icNL
PS
u
ite
in
tr
o
d
u
ce
d
lar
g
e
-
s
ca
le
co
r
p
o
r
a,
p
r
etr
ain
e
d
m
o
d
els,
an
d
ev
alu
atio
n
b
en
ch
m
ar
k
s
,
s
ig
n
if
ican
tly
ad
v
an
cin
g
r
esear
ch
in
u
n
d
er
-
r
eso
u
r
ce
d
I
n
d
ic
lan
g
u
ag
e
p
r
o
c
ess
in
g
[
1
6
]
.
T
o
ex
p
lo
r
e
th
e
af
o
r
em
en
tio
n
ed
ca
p
a
b
ilit
ies
o
f
ML
Ms,
th
e
p
r
o
ject
is
d
i
v
id
ed
in
to
two
co
r
e
task
s
:
ev
alu
atin
g
cr
o
s
s
-
lin
g
u
al
s
em
an
tic
s
im
ilar
ity
u
s
in
g
wo
r
d
em
b
ed
d
in
g
s
f
r
o
m
m
o
d
els
lik
e
B
L
OOM
an
d
L
aBS
E
[
1
7
]
,
an
d
test
in
g
th
e
ze
r
o
-
s
h
o
t
in
f
er
en
ce
ab
ilit
y
o
f
ML
Ms th
r
o
u
g
h
a
n
atu
r
al
lan
g
u
a
g
e
in
f
er
e
n
ce
(
NL
I
)
task
.
B
y
lev
er
ag
in
g
p
r
etr
ain
e
d
m
o
d
els,
cu
r
ated
d
atasets
,
an
d
v
is
u
aliza
tio
n
s
,
th
e
s
tu
d
y
aim
s
to
ass
ess
th
e
ef
f
ec
tiv
en
ess
an
d
r
e
a
l
-
w
o
r
l
d
a
p
p
l
i
c
a
b
i
li
t
y
o
f
M
L
M
s
i
n
m
u
l
t
i
li
n
g
u
a
l
s
e
t
t
i
n
g
s
,
p
a
r
t
i
c
u
l
a
r
l
y
f
o
r
l
o
w
-
r
es
o
u
r
c
e
la
n
g
u
a
g
e
s
.
Alth
o
u
g
h
MLMs
ar
e
tr
ain
ed
o
n
lar
g
e
an
d
d
iv
er
s
e
co
r
p
o
r
a,
th
eir
ab
ilit
y
to
r
ep
r
esen
t
s
em
an
tic
m
ea
n
in
g
co
n
s
is
ten
tly
ac
r
o
s
s
lan
g
u
ag
es
is
n
o
t
g
u
ar
a
n
teed
,
esp
ec
ially
f
o
r
lo
w
-
r
eso
u
r
ce
la
n
g
u
ag
es.
Prio
r
s
tu
d
ies
h
av
e
s
h
o
wn
th
at
m
u
ltil
in
g
u
a
l
alig
n
m
en
t
v
ar
ies
s
ig
n
if
ican
tly
d
ep
en
d
i
n
g
o
n
m
o
d
el
ar
c
h
itectu
r
e,
tr
ain
in
g
o
b
jectiv
es,
an
d
lan
g
u
ag
e
ch
a
r
ac
ter
is
tics
.
Ho
wev
er
,
m
a
n
y
wo
r
k
s
eith
er
f
o
c
u
s
ex
clu
s
iv
el
y
o
n
h
ig
h
-
r
eso
u
r
ce
lan
g
u
ag
es
o
r
ev
alu
ate
m
o
d
els
u
s
in
g
a
s
in
g
le
task
.
I
n
d
ic
lan
g
u
ag
es
s
u
ch
as
Kan
n
ad
a
r
em
ai
n
u
n
d
er
ex
p
lo
r
ed
in
cr
o
s
s
-
lin
g
u
al
ev
alu
atio
n
.
T
h
is
wo
r
k
ad
d
r
ess
es
th
is
g
ap
b
y
jo
in
tly
an
aly
zin
g
wo
r
d
-
lev
el
s
em
an
tic
s
im
ilar
ity
an
d
s
en
ten
ce
-
le
v
el
in
f
er
e
n
ce
ac
r
o
s
s
m
u
ltip
le
lan
g
u
ag
es.
B
y
co
m
p
ar
i
n
g
g
en
er
al
-
p
u
r
p
o
s
e
m
u
ltil
in
g
u
al
m
o
d
els
with
task
-
s
p
ec
if
ic
ar
ch
itectu
r
es,
we
aim
to
b
etter
u
n
d
er
s
tan
d
th
e
s
tr
en
g
t
h
s
an
d
lim
itatio
n
s
o
f
cu
r
r
en
t
m
u
ltil
in
g
u
al
m
o
d
els in
lo
w
-
r
e
s
o
u
r
ce
I
n
d
ic
lan
g
u
a
g
e
s
ettin
g
s
.
T
h
e
p
r
im
ar
y
o
b
jectiv
e
o
f
th
i
s
wo
r
k
is
to
ev
alu
ate
h
o
w
e
f
f
ec
tiv
ely
ML
Ms
ca
p
tu
r
e
cr
o
s
s
-
lin
g
u
al
s
em
an
tic
alig
n
m
en
t
a
n
d
tr
an
s
f
er
th
is
k
n
o
wled
g
e
to
d
o
wn
s
tr
ea
m
task
s
.
T
o
ac
h
iev
e
th
is
,
w
e
p
r
esen
t
a
u
n
i
f
ied
ev
alu
atio
n
f
r
am
ewo
r
k
th
at
an
a
ly
ze
s
b
o
th
wo
r
d
-
lev
el
s
em
an
ti
c
s
im
ilar
ity
an
d
s
en
ten
ce
-
lev
el
ze
r
o
-
s
h
o
t
tr
a
n
s
f
er
p
er
f
o
r
m
an
ce
ac
r
o
s
s
E
n
g
lis
h
,
Fre
n
ch
,
Hin
d
i,
an
d
Kan
n
a
d
a.
W
e
co
m
p
ar
e
g
en
er
al
-
p
u
r
p
o
s
e
ML
Ms
s
u
ch
as
B
L
OOM
with
task
-
s
p
ec
ialize
d
m
o
d
els
in
clu
d
in
g
L
aBS
E
a
n
d
XL
M
-
R
to
s
tu
d
y
th
e
im
p
ac
t
o
f
a
r
ch
itectu
r
al
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
C
r
o
s
s
-
lin
g
u
a
l sema
n
tic
a
lig
n
men
t a
n
d
tr
a
n
s
fer lea
r
n
in
g
u
s
in
g
mu
ltil
in
g
u
a
l la
n
g
u
a
g
e
mo
d
els
(
N
ir
a
n
ja
n
G
C
)
975
d
esig
n
an
d
tr
ain
in
g
o
b
jectiv
e
s
o
n
cr
o
s
s
-
lin
g
u
al
p
e
r
f
o
r
m
an
ce
.
Ou
r
ex
p
er
im
e
n
ts
p
r
o
v
id
e
em
p
ir
ical
ev
id
e
n
ce
th
at
s
tr
o
n
g
wo
r
d
-
lev
el
alig
n
m
en
t
d
o
es
n
o
t
n
ec
ess
ar
ily
lead
to
im
p
r
o
v
e
d
s
en
ten
ce
-
lev
el
in
f
er
en
ce
,
p
ar
ticu
lar
ly
in
lo
w
-
r
eso
u
r
ce
la
n
g
u
a
g
es su
ch
as Ka
n
n
ad
a.
2.
M
E
T
H
O
D
T
h
e
s
y
s
tem
ar
ch
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
s
o
lu
tio
n
is
d
esig
n
e
d
to
f
ac
ilit
ate
cr
o
s
s
-
lin
g
u
al
ev
al
u
atio
n
o
f
ML
Ms
th
r
o
u
g
h
two
co
r
e
m
o
d
u
les:
wo
r
d
em
b
ed
d
in
g
s
i
m
ilar
ity
an
d
NL
I
.
I
t
b
u
ild
s
o
n
th
e
m
eth
o
d
o
lo
g
y
p
r
o
p
o
s
ed
in
[
1
]
,
b
u
t
with
f
o
c
u
s
o
n
I
n
d
ic
lan
g
u
ag
es
esp
ec
i
ally
Kan
n
ad
a.
E
ac
h
m
o
d
u
le
s
u
p
p
o
r
ts
a
s
p
ec
if
ic
d
o
wn
s
tr
ea
m
task
an
d
r
elies
o
n
p
r
e
-
t
r
ain
ed
t
r
an
s
f
o
r
m
e
r
-
b
as
ed
m
o
d
els
to
m
ea
s
u
r
e
m
u
ltil
in
g
u
al
alig
n
m
en
t
a
n
d
s
em
an
tic
u
n
d
er
s
tan
d
in
g
.
T
h
e
in
ter
f
ac
e
ac
ts
as
th
e
u
s
er
in
ter
ac
tio
n
lay
er
,
w
h
ile
b
ac
k
en
d
lo
g
ic
h
a
n
d
les
tr
an
s
latio
n
,
em
b
ed
d
in
g
g
en
er
a
tio
n
,
an
d
in
f
er
e
n
ce
u
s
in
g
s
tr
ea
m
lin
ed
wo
r
k
f
lo
ws.
Fig
u
r
e
1
d
ep
icts
th
e
s
y
s
tem
ar
ch
itectu
r
e
o
f
t
h
e
ap
p
licati
o
n
.
I
n
t
h
e
wo
r
d
em
b
ed
d
in
g
s
im
ilar
ity
m
o
d
u
le,
th
e
u
s
er
b
eg
i
n
s
b
y
en
ter
in
g
an
E
n
g
lis
h
wo
r
d
t
h
r
o
u
g
h
th
e
in
ter
f
ac
e.
T
h
e
s
y
s
tem
r
etr
iev
es
th
e
co
r
r
esp
o
n
d
in
g
tr
an
s
latio
n
s
in
Fre
n
ch
,
Hin
d
i,
a
n
d
Kan
n
ad
a
u
s
in
g
a
tr
an
s
latio
n
u
tili
ty
.
T
h
e
w
o
r
d
an
d
its
tr
an
s
latio
n
s
ar
e
to
k
en
ized
,
an
d
a
p
r
elo
ad
e
d
ML
M
(
L
aBS
E
)
g
en
er
ates
v
ec
to
r
em
b
ed
d
in
g
s
f
o
r
ea
ch
wo
r
d
.
T
h
ese
em
b
ed
d
in
g
s
ar
e
co
m
p
ar
ed
u
s
in
g
p
air
wis
e
co
s
in
e
s
im
ilar
ity
to
m
ea
s
u
r
e
s
em
an
tic
p
r
o
x
im
ity
ac
r
o
s
s
lan
g
u
ag
es
[
1
8
]
.
T
h
e
s
im
ilar
ity
s
co
r
es
ar
e
ac
co
m
p
an
ied
b
y
v
is
u
aliza
tio
n
o
u
t
p
u
ts
in
clu
d
in
g
h
ea
tm
a
p
s
an
d
t
-
SNE
p
lo
ts
,
p
r
o
v
id
i
n
g
in
s
ig
h
t
s
in
to
th
e
d
eg
r
ee
o
f
alig
n
m
en
t
b
etwe
en
cr
o
s
s
-
lin
g
u
al
wo
r
d
r
e
p
r
esen
tatio
n
s
.
Fig
u
r
e
1
.
Sy
s
tem
a
r
c
h
itectu
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
2
,
Ap
r
il
20
2
6
:
9
7
3
-
980
976
I
n
th
e
NL
I
m
o
d
u
le,
th
e
u
s
er
i
n
p
u
ts
a
p
r
em
is
e
an
d
a
h
y
p
o
th
es
is
,
b
o
th
o
f
wh
ich
ar
e
f
ir
s
t
p
r
ep
r
o
ce
s
s
ed
.
T
h
e
p
r
ep
r
o
ce
s
s
in
g
in
v
o
lv
es
cl
ea
n
in
g
t
h
e
d
ata,
tr
u
n
ca
tin
g
o
r
p
ad
d
in
g
th
e
in
p
u
t
to
a
f
ix
ed
le
n
g
th
o
f
1
2
8
to
k
en
s
,
an
d
c
o
n
v
e
r
tin
g
th
e
to
k
en
s
i
n
to
ten
s
o
r
s
al
o
n
g
with
atten
tio
n
m
ask
s
.
T
h
is
f
o
r
m
atted
i
n
p
u
t
is
f
ed
in
to
a
f
in
e
-
tu
n
ed
XL
M
-
R
m
o
d
el
[
1
9
]
,
wh
ich
h
as
b
ee
n
tr
ain
ed
o
n
t
h
e
E
n
g
lis
h
XNL
I
d
ataset
an
d
ev
alu
ated
o
n
th
e
Fre
n
c
h
XNL
I
test
d
atase
t
[
2
0
]
an
d
Kan
n
ad
a
I
n
d
icXNL
I
test
d
ataset
[
2
1
]
.
T
h
e
m
o
d
el
p
r
o
ce
s
s
es th
e
p
r
ep
r
o
ce
s
s
ed
in
p
u
t
an
d
class
if
ies
th
e
r
elatio
n
s
h
ip
b
etwe
en
th
e
p
r
em
is
e
an
d
h
y
p
o
th
esis
in
to
o
n
e
o
f
th
r
ee
ca
teg
o
r
ies:
en
tailm
en
t,
co
n
tr
ad
ictio
n
,
o
r
n
eu
tr
al.
T
h
e
r
esu
ltin
g
p
r
e
d
ictio
n
an
d
c
o
n
f
id
en
ce
s
co
r
es
ar
e
d
is
p
lay
ed
to
th
e
u
s
er
i
n
a
r
ea
d
ab
le
f
o
r
m
at.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
is
s
ec
tio
n
p
r
esen
ts
th
e
r
es
u
lts
o
b
tain
ed
th
r
o
u
g
h
a
s
er
ie
s
o
f
ex
p
e
r
im
en
ts
co
n
d
u
cted
o
n
v
a
r
io
u
s
ML
Ms
ac
r
o
s
s
th
e
two
m
o
d
u
les:
w
o
r
d
em
b
ed
d
i
n
g
s
im
ilar
ity
an
d
NL
I
.
I
t
p
r
o
v
id
es
a
d
eta
iled
ac
co
u
n
t
o
f
th
e
p
er
f
o
r
m
an
ce
o
f
ea
ch
m
o
d
el,
s
u
p
p
o
r
ted
b
y
r
elev
a
n
t
tab
les
an
d
v
is
u
aliza
tio
n
s
.
T
h
ese
ev
alu
atio
n
s
o
f
f
er
in
s
ig
h
ts
in
to
th
e
s
tr
en
g
th
s
a
n
d
lim
itatio
n
s
o
f
ea
c
h
m
o
d
el
in
h
a
n
d
lin
g
m
u
ltil
in
g
u
al
task
s
an
d
tr
an
s
f
er
r
in
g
k
n
o
wled
g
e
ac
r
o
s
s
h
ig
h
-
an
d
lo
w
-
r
eso
u
r
ce
lan
g
u
ag
es.
3
.
1
.
Wo
rd
e
m
bedd
ing
s
im
ila
rit
y
re
s
ults
Fo
r
th
e
wo
r
d
em
b
ed
d
i
n
g
s
im
il
ar
ity
m
o
d
u
le,
th
e
m
o
d
els
ev
al
u
ated
in
clu
d
e
B
L
OOM
5
6
0
M,
B
L
OOM
1
.
7
B
,
an
d
L
aBS
E
.
T
h
eir
ef
f
ec
tiv
en
ess
is
m
ea
s
u
r
ed
u
s
in
g
th
e
av
er
ag
e
p
air
wis
e
co
s
in
e
s
im
ilar
ity
s
co
r
e
(
1
)
co
m
p
u
ted
o
v
e
r
a
d
ataset
o
f
3
0
0
0
E
n
g
lis
h
wo
r
d
s
an
d
th
eir
tr
an
s
latio
n
s
in
Fre
n
ch
,
Hin
d
i,
a
n
d
Kan
n
a
d
a,
wh
ich
q
u
an
tifie
s
th
e
s
em
an
tic
alig
n
m
en
t b
etwe
en
wo
r
d
em
b
e
d
d
in
g
s
ac
r
o
s
s
d
if
f
er
en
t la
n
g
u
a
g
es.
(
⃗
,
⃗
⃗
)
=
⃗
⋅
⃗
⃗
∥
⃗
∥
∥
⃗
⃗
∥
(
1
)
w
h
er
e
⃗
a
n
d
⃗
⃗
ar
e
th
e
wo
r
d
em
b
ed
d
in
g
v
ec
to
r
s
,
⃗
⋅
⃗
⃗
is
th
e
d
o
t
p
r
o
d
u
ct
,
an
d
∥
⃗
∥
∥
⃗
⃗
∥
ar
e
th
e
m
ag
n
itu
d
es
o
f
th
e
v
ec
to
r
s
.
T
ab
le
1
p
r
esen
ts
th
e
av
er
ag
e
p
air
wis
e
co
s
in
e
s
im
ilar
ity
s
co
r
es
o
b
tain
ed
b
y
th
e
th
r
ee
ML
Ms
-
B
L
OOM
5
6
0
M
,
B
L
OOM
1
.
7
B
,
an
d
L
aBS
E
ac
r
o
s
s
d
i
f
f
er
en
t
lan
g
u
ag
e
p
air
s
in
th
e
wo
r
d
em
b
ed
d
in
g
s
im
ilar
ity
m
o
d
u
le.
T
ab
le
1
.
Av
e
r
ag
e
p
air
wis
e
co
s
in
e
s
im
ilar
ity
s
co
r
es
La
n
g
u
a
g
e
P
a
i
r
B
LO
O
M
5
6
0
M
B
LO
O
M
1
.
7
B
La
B
S
E
EN
-
FR
0
.
9
8
0
.
9
5
0
.
9
0
EN
-
HI
-
0
.
0
0
3
-
0
.
1
4
0
.
8
9
EN
-
KN
-
0
.
1
6
-
0
.
1
8
0
.
8
5
H
I
-
KN
0
.
0
9
7
0
.
0
3
2
0
.
8
2
T
h
e
co
m
p
a
r
ativ
ely
h
ig
h
er
s
im
ilar
ity
s
co
r
es
ac
h
iev
ed
b
y
L
aBS
E
ac
r
o
s
s
all
lan
g
u
ag
e
p
air
s
i
n
d
icate
its
s
u
p
er
io
r
ab
ilit
y
to
ca
p
tu
r
e
s
em
an
tic
alig
n
m
en
t
ac
r
o
s
s
lan
g
u
ag
es
an
d
was
th
u
s
ch
o
s
en
as
th
e
m
o
s
t
s
u
itab
le
m
o
d
el
f
o
r
d
ep
l
o
y
m
en
t
i
n
th
e
W
o
r
d
E
m
b
ed
d
in
g
Similar
ity
m
o
d
u
le.
R
ec
en
t
s
tu
d
ies
h
av
e
em
p
h
asized
th
at
cr
o
s
s
-
lin
g
u
al
alig
n
m
en
t
is
n
o
t
a
f
ix
ed
p
r
o
p
er
ty
o
f
m
u
ltil
in
g
u
al
m
o
d
els
b
u
t
ca
n
b
e
i
n
f
lu
en
ce
d
b
y
ar
c
h
itectu
r
al
an
d
tr
ain
i
n
g
s
tr
ateg
ies.
T
ec
h
n
iq
u
es
s
u
ch
as
c
r
o
s
s
-
lin
g
u
al
p
o
s
itio
n
en
co
d
in
g
an
d
b
id
ir
ec
t
io
n
al
tr
ain
in
g
h
a
v
e
b
ee
n
s
h
o
wn
to
im
p
r
o
v
e
alig
n
m
en
t
b
y
e
x
p
licitly
m
o
d
elin
g
wo
r
d
o
r
d
er
d
if
f
er
e
n
ce
s
ac
r
o
s
s
lan
g
u
ag
es
an
d
s
tr
en
g
th
en
in
g
b
ilin
g
u
al
r
ep
r
esen
tatio
n
s
[
2
2
]
,
[
2
3
]
.
I
n
co
n
tr
a
s
t,
th
e
p
r
esen
t
wo
r
k
e
v
alu
ates
p
r
etr
ain
e
d
m
o
d
els
with
o
u
t
alig
n
m
e
n
t
s
p
ec
if
ic
f
in
e
tu
n
i
n
g
.
T
h
e
o
b
s
er
v
ed
d
eg
r
ad
atio
n
in
alig
n
m
en
t
f
o
r
I
n
d
ic
lan
g
u
ag
es,
p
ar
ticu
lar
ly
in
B
L
OOM
em
b
ed
d
in
g
s
,
s
u
g
g
ests
th
at
in
co
r
p
o
r
atin
g
s
u
c
h
alig
n
m
e
n
t
awa
r
e
s
tr
ateg
ies
co
u
ld
p
o
ten
tially
en
h
a
n
ce
cr
o
s
s
lin
g
u
al
r
o
b
u
s
tn
ess
.
An
o
th
er
im
p
o
r
tan
t
f
ac
to
r
af
f
ec
tin
g
p
er
f
o
r
m
a
n
ce
in
lo
w
r
eso
u
r
ce
lan
g
u
ag
es
is
lex
ical
f
r
eq
u
en
c
y
.
Prio
r
wo
r
k
h
as
s
h
o
wn
t
h
at
m
u
ltil
in
g
u
al
tr
ain
in
g
an
d
k
n
o
wled
g
e
d
is
till
atio
n
ten
d
to
f
av
o
r
h
i
g
h
f
r
eq
u
e
n
cy
wo
r
d
s
,
lead
in
g
to
wea
k
er
r
ep
r
esen
tatio
n
s
f
o
r
lo
w
f
r
e
q
u
en
c
y
lex
ical
item
s
[
2
4
]
,
lik
ely
co
n
tr
ib
u
tin
g
to
t
h
e
r
ed
u
ce
d
p
er
f
o
r
m
an
ce
o
b
s
er
v
e
d
f
o
r
Kan
n
ad
a.
Fig
u
r
e
2
p
r
esen
ts
th
e
em
b
ed
d
in
g
v
is
u
aliza
tio
n
s
.
Fig
u
r
e
s
2
(
a)
,
2
(
b
)
,
an
d
2
(
c
)
p
r
esen
t
th
e
t
-
SNE
v
is
u
aliza
tio
n
o
f
th
e
v
ec
to
r
e
m
b
ed
d
in
g
s
g
en
er
ated
b
y
th
e
B
L
OOM
5
6
0
M,
B
L
OOM
1
.
7
B
,
an
d
th
e
L
aBS
E
m
o
d
el
r
esp
ec
tiv
el
y
,
in
a
two
-
d
im
en
s
io
n
al
s
p
ac
e.
T
h
e
s
ig
n
if
ic
an
t
o
v
e
r
lap
am
o
n
g
th
e
e
m
b
ed
d
in
g
s
o
f
tr
an
s
lated
wo
r
d
s
co
n
f
ir
m
s
th
at
L
aBS
E
ef
f
ec
tiv
ely
ca
p
tu
r
es
th
e
s
em
an
tic
alig
n
m
en
t
o
f
r
elate
d
w
o
r
d
s
ac
r
o
s
s
d
if
f
er
e
n
t
lan
g
u
ag
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
C
r
o
s
s
-
lin
g
u
a
l sema
n
tic
a
lig
n
men
t a
n
d
tr
a
n
s
fer lea
r
n
in
g
u
s
in
g
mu
ltil
in
g
u
a
l la
n
g
u
a
g
e
mo
d
els
(
N
ir
a
n
ja
n
G
C
)
977
(
a)
(
b
)
(
c)
Fig
u
r
e
2
.
t
-
SNE
v
is
u
aliza
tio
n
o
f
em
b
e
d
d
in
g
s
o
b
tain
ed
f
r
o
m
th
e
ML
Ms (
a)
B
L
OOM
5
6
0
M
,
(
b
)
B
L
OOM
1
.
7
B
,
an
d
(
c)
L
aBS
E
3
.
2
.
NL
I
r
esu
lt
s
Fo
r
th
e
NL
I
m
o
d
u
le,
t
h
e
ze
r
o
-
s
h
o
t
tr
an
s
f
er
lear
n
in
g
p
er
f
o
r
m
an
ce
o
f
B
L
OOM
5
6
0
M,
B
L
OOM
1
.
7
B
,
an
d
XL
M
-
R
is
co
m
p
ar
ed
u
s
in
g
s
tan
d
ar
d
class
if
icatio
n
m
e
tr
ics
-
a
cc
u
r
ac
y
an
d
F1
s
co
r
e
o
n
test
d
atasets
in
Fre
n
ch
an
d
Kan
n
a
d
a
.
T
ab
le
2
s
u
m
m
ar
izes
th
e
test
ac
cu
r
ac
y
s
co
r
es
an
d
T
ab
le
3
s
u
m
m
ar
izes
th
e
test
F1
s
co
r
es
f
o
r
th
e
NL
I
task
,
o
b
tain
ed
b
y
ev
alu
atin
g
th
e
th
r
ee
af
o
r
e
m
en
tio
n
ed
f
in
e
-
tu
n
e
d
ML
Ms
o
n
Fre
n
ch
a
n
d
Kan
n
ad
a
test
d
atasets
.
XL
M
-
R
ac
h
iev
es
b
etter
test
ac
cu
r
ac
y
an
d
F1
s
co
r
es
f
o
r
b
o
t
h
Fre
n
c
h
an
d
Ka
n
n
ad
a,
in
d
icatin
g
its
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
in
ze
r
o
-
s
h
o
t
tr
a
n
s
f
er
f
o
r
th
e
NL
I
task
.
C
o
n
s
eq
u
en
tly
,
XL
M
-
R
was
s
elec
ted
f
o
r
d
ep
lo
y
m
e
n
t
in
th
e
NL
I
m
o
d
u
le.
T
h
e
s
u
p
er
io
r
NL
I
p
er
f
o
r
m
an
ce
o
f
XL
M
-
R
ca
n
also
b
e
attr
ib
u
ted
to
its
ar
ch
itectu
r
e
an
d
p
r
etr
ain
in
g
s
tr
ateg
y
,
wh
ich
ar
e
b
etter
s
u
it
ed
f
o
r
s
en
ten
ce
le
v
el
r
ea
s
o
n
i
n
g
.
C
r
o
s
s
-
atten
tio
n
m
ec
h
an
is
m
s
p
lay
a
cr
itical
r
o
l
e
in
e
n
ab
lin
g
alig
n
m
e
n
t
b
etwe
en
p
r
e
m
is
e
an
d
h
y
p
o
th
esis
r
ep
r
esen
tatio
n
s
ac
r
o
s
s
lan
g
u
ag
es.
Prio
r
a
n
aly
s
es
h
av
e
id
en
tifie
d
lim
itatio
n
s
in
cr
o
s
s
-
atten
tio
n
b
eh
av
i
o
r
th
at
af
f
ec
t
co
n
tex
tu
al
alig
n
m
en
t,
p
a
r
ticu
lar
ly
in
n
o
n
-
au
to
r
e
g
r
ess
iv
e
an
d
m
u
ltil
in
g
u
al
s
ettin
g
s
[
2
5
]
.
Alth
o
u
g
h
th
is
wo
r
k
d
o
es
n
o
t
p
er
f
o
r
m
atten
tio
n
-
le
v
el
an
aly
s
is
,
th
e
co
n
s
is
ten
t
im
p
r
o
v
em
e
n
t
o
f
XL
M
-
R
o
v
er
B
L
OOM
in
d
icate
s
th
at
s
tr
o
n
g
er
an
d
m
o
r
e
co
n
te
x
t
-
awa
r
e
cr
o
s
s
-
atten
tio
n
r
ep
r
esen
tatio
n
s
ar
e
b
en
ef
icial
f
o
r
cr
o
s
s
-
lin
g
u
al
in
f
er
en
ce
.
F
ig
u
r
e
3
p
r
esen
ts
th
e
m
o
d
el
wis
e
ac
cu
r
ac
y
an
d
F1
s
co
r
e
co
m
p
ar
is
o
n
s
.
Fig
u
r
e
3
(
a
)
p
r
e
s
en
ts
a
b
ar
ch
ar
t
o
f
t
h
e
test
ac
cu
r
ac
y
s
co
r
es,
an
d
Fig
u
r
e
3
(
b
)
d
e
p
icts
a
b
ar
ch
a
r
t
o
f
th
e
test
F1
s
co
r
es
f
o
r
th
e
NL
I
task
,
o
b
tain
ed
b
y
ev
al
u
atin
g
th
e
th
r
ee
f
in
e
-
tu
n
ed
ML
Ms
-
B
L
OOM
5
6
0
M,
B
L
OOM
1
.
7
B
,
XL
M
-
R
o
n
Fre
n
ch
an
d
Kan
n
ad
a
test
d
atasets
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
2
,
Ap
r
il
20
2
6
:
9
7
3
-
980
978
T
ab
le
2
.
T
est ac
cu
r
ac
y
s
co
r
es
La
n
g
u
a
g
e
B
LO
O
M
5
6
0
M
B
LO
O
M
1
.
7
B
X
LM
-
R
F
r
e
n
c
h
7
2
.
2
7
%
7
5
.
5
1
%
7
4
.
8
3
%
K
a
n
n
a
d
a
5
9
.
0
8
%
6
1
.
5
4
%
6
8
.
8
%
T
ab
le
3
.
T
est F1
s
co
r
es
La
n
g
u
a
g
e
B
LO
O
M
5
6
0
M
B
LO
O
M
1
.
7
B
X
LM
-
R
F
r
e
n
c
h
7
2
.
4
%
7
5
.
5
9
%
7
4
.
8
1
%
K
a
n
n
a
d
a
5
8
.
9
8
%
6
1
.
4
6
%
6
8
.
6
2
%
(
a)
(
b
)
Fig
u
r
e
3
.
B
ar
ch
a
r
t f
o
r
(
a)
test
ac
cu
r
ac
y
s
co
r
es a
n
d
(
b
)
test
F1
s
co
r
es
4.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
s
h
o
ws
th
at
cr
o
s
s
-
lin
g
u
al
p
er
f
o
r
m
an
ce
o
f
ML
Ms
is
h
ig
h
ly
task
-
d
ep
en
d
en
t
a
n
d
ca
n
n
o
t
b
e
in
f
er
r
ed
s
o
lely
f
r
o
m
m
o
d
el
s
ize
o
r
m
u
ltil
in
g
u
al
c
o
v
er
a
g
e.
Ou
r
ex
p
er
im
e
n
ts
d
em
o
n
s
tr
ate
th
at
task
-
s
p
ec
ialized
m
o
d
els
s
u
ch
as
L
aBS
E
an
d
XL
M
-
R
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
g
e
n
er
al
p
u
r
p
o
s
e
m
u
ltil
in
g
u
al
m
o
d
els
lik
e
B
L
OOM
f
o
r
wo
r
d
lev
el
s
em
a
n
tic
s
im
ilar
ity
an
d
s
en
ten
ce
le
v
el
in
f
er
e
n
ce
,
p
a
r
ticu
lar
ly
in
l
o
w
r
eso
u
r
ce
I
n
d
ic
lan
g
u
ag
es
s
u
ch
as
Kan
n
ad
a.
T
h
ese
f
in
d
in
g
s
h
ig
h
lig
h
t
th
e
im
p
o
r
tan
ce
o
f
ar
ch
itectu
r
al
d
esig
n
an
d
tr
ain
in
g
o
b
jectiv
es
wh
en
s
elec
tin
g
m
u
ltil
in
g
u
al
m
o
d
els
f
o
r
p
r
ac
ti
ca
l
ap
p
licatio
n
s
.
W
h
ile
th
is
wo
r
k
f
o
c
u
s
es
o
n
ev
alu
atin
g
p
r
etr
ain
ed
m
o
d
els
,
f
u
tu
r
e
w
o
r
k
c
o
u
ld
e
x
p
lo
r
e
tr
ain
in
g
s
tr
ateg
ies
s
u
ch
as
p
r
o
g
r
ess
iv
e
m
u
lti
-
g
r
an
u
lar
ity
lear
n
in
g
,
w
h
er
e
m
o
d
els ar
e
tr
ain
ed
f
r
o
m
wo
r
d
s
to
p
h
r
ases
an
d
s
en
ten
ce
s
,
to
f
u
r
th
er
im
p
r
o
v
e
cr
o
s
s
-
lin
g
u
al
r
o
b
u
s
tn
ess
an
d
tr
an
s
f
e
r
p
er
f
o
r
m
a
n
ce
in
lo
w
r
eso
u
r
c
e
s
ettin
g
s
.
I
n
ad
d
itio
n
,
f
u
tu
r
e
en
h
an
ce
m
e
n
ts
m
ay
in
clu
d
e
in
v
esti
g
atin
g
o
n
e
-
s
h
o
t
an
d
f
ew
-
s
h
o
t tr
an
s
f
er
ca
p
a
b
ilit
ies,
ca
p
tu
r
in
g
s
en
ten
ce
-
lev
el
s
em
an
tic
alig
n
m
en
t
m
o
r
e
ex
p
licitly
,
an
d
e
x
ten
d
in
g
th
e
an
aly
s
is
to
o
th
er
d
o
wn
s
tr
e
am
NL
P task
s
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
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
Nir
an
jan
G
C
✓
✓
✓
✓
✓
✓
✓
✓
✓
R
am
ak
an
th
Ku
m
ar
P
✓
✓
✓
✓
✓
✓
✓
Pav
ith
r
a
H
✓
✓
✓
✓
Min
al
Mo
h
ar
ir
✓
✓
✓
✓
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
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
C
r
o
s
s
-
lin
g
u
a
l sema
n
tic
a
lig
n
men
t a
n
d
tr
a
n
s
fer lea
r
n
in
g
u
s
in
g
mu
ltil
in
g
u
a
l la
n
g
u
a
g
e
mo
d
els
(
N
ir
a
n
ja
n
G
C
)
979
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
au
th
o
r
s
co
n
f
ir
m
th
at
th
e
d
ata
s
u
p
p
o
r
tin
g
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
av
ailab
le
with
in
th
e
ar
ticle
.
RE
F
E
R
E
NC
E
S
[
1
]
S
.
K
a
k
a
r
l
a
,
G
.
S
.
B
.
V
e
n
k
a
t
a
,
a
n
d
A
.
G
a
d
d
a
m,
“
H
o
w
d
o
e
s
a
m
u
l
t
i
l
i
n
g
u
a
l
L
M
h
a
n
d
l
e
mu
l
t
i
p
l
e
l
a
n
g
u
a
g
e
s
?
,
”
a
r
Xi
v
p
re
p
ri
n
t
a
rXi
v
:
2
5
0
2
.
0
4
2
6
9
,
2
0
2
5
.
[
2
]
S
.
R
a
j
a
e
e
a
n
d
C
.
M
o
n
z
,
“
A
n
a
l
y
z
i
n
g
t
h
e
e
v
a
l
u
a
t
i
o
n
o
f
c
r
o
ss
-
l
i
n
g
u
a
l
k
n
o
w
l
e
d
g
e
t
r
a
n
sf
e
r
i
n
m
u
l
t
i
l
i
n
g
u
a
l
l
a
n
g
u
a
g
e
m
o
d
e
l
s,”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
1
8
t
h
C
o
n
f
e
re
n
c
e
o
f
t
h
e
Eu
r
o
p
e
a
n
C
h
a
p
t
e
r
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
s
t
i
c
s
(
V
o
l
u
m
e
1
:
L
o
n
g
Pa
p
e
rs)
,
2
0
2
4
,
p
p
.
2
8
9
5
–
2
9
1
4
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
4
.
e
a
c
l
-
l
o
n
g
.
1
7
7
.
[
3
]
R
.
B
a
w
d
e
n
a
n
d
F
.
Y
v
o
n
,
“
I
n
v
e
st
i
g
a
t
i
n
g
t
h
e
t
r
a
n
sl
a
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
o
f
a
l
a
r
g
e
mu
l
t
i
l
i
n
g
u
a
l
l
a
n
g
u
a
g
e
m
o
d
e
l
:
t
h
e
c
a
s
e
o
f
b
l
o
o
m
,
”
a
rXi
v
p
re
p
r
i
n
t
a
rXi
v
:
2
3
0
3
.
0
1
9
1
1
,
2
0
2
3
.
[
4
]
Z.
C
h
i
e
t
a
l
.
,
“
X
L
M
-
E:
c
r
o
ss
-
l
i
n
g
u
a
l
l
a
n
g
u
a
g
e
m
o
d
e
l
p
r
e
-
t
r
a
i
n
i
n
g
v
i
a
E
LE
C
TR
A
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
6
0
t
h
A
n
n
u
a
l
M
e
e
t
i
n
g
o
f
t
h
e
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
(
V
o
l
u
m
e
1
:
L
o
n
g
Pa
p
e
r
s)
,
2
0
2
2
,
p
p
.
6
1
7
0
–
6
1
8
2
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
2
.
a
c
l
-
l
o
n
g
.
4
2
7
.
[
5
]
J.
H
u
,
S
.
R
u
d
e
r
,
A
.
S
i
d
d
h
a
n
t
,
G
.
N
e
u
b
i
g
,
O
.
F
i
r
a
t
,
a
n
d
M
.
J
o
h
n
s
o
n
,
“
X
t
r
e
me:
a
ma
ssi
v
e
l
y
m
u
l
t
i
l
i
n
g
u
a
l
m
u
l
t
i
-
t
a
sk
b
e
n
c
h
m
a
r
k
f
o
r
e
v
a
l
u
a
t
i
n
g
c
r
o
ss
-
l
i
n
g
u
a
l
g
e
n
e
r
a
l
i
s
a
t
i
o
n
,
”
i
n
I
n
t
e
rn
a
t
i
o
n
a
l
c
o
n
f
e
re
n
c
e
o
n
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
,
2
0
2
0
,
p
p
.
4
4
1
1
–
4
4
2
1
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
0
0
3
.
1
1
0
8
0
.
[
6
]
S
.
W
u
a
n
d
M
.
D
r
e
d
z
e
,
“
A
r
e
a
l
l
l
a
n
g
u
a
g
e
s
c
r
e
a
t
e
d
e
q
u
a
l
i
n
m
u
l
t
i
l
i
n
g
u
a
l
B
ER
T
?
,
”
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
A
n
n
u
a
l
Me
e
t
i
n
g
o
f
t
h
e
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
,
n
o
.
J
u
l
y
,
p
p
.
1
2
0
–
1
3
0
,
2
0
2
0
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
r
e
p
l
4
n
l
p
-
1
.
1
6
.
[
7
]
Z.
W
a
n
g
,
Z.
C
.
L
i
p
t
o
n
,
a
n
d
Y
.
Ts
v
e
t
k
o
v
,
“
O
n
n
e
g
a
t
i
v
e
i
n
t
e
r
f
e
r
e
n
c
e
i
n
m
u
l
t
i
l
i
n
g
u
a
l
mo
d
e
l
s:
f
i
n
d
i
n
g
s
a
n
d
a
m
e
t
a
-
l
e
a
r
n
i
n
g
t
r
e
a
t
me
n
t
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
2
0
C
o
n
f
e
re
n
c
e
o
n
Em
p
i
r
i
c
a
l
M
e
t
h
o
d
s
i
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pro
c
e
ss
i
n
g
(
E
M
N
L
P)
,
2
0
2
0
,
p
p
.
4
4
3
8
–
4
4
5
0
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
e
mn
l
p
-
ma
i
n
.
3
5
9
.
[
8
]
R
.
C
h
o
e
n
n
i
,
D
.
G
a
r
r
e
t
t
e
,
a
n
d
E.
S
h
u
t
o
v
a
,
“
C
r
o
ss
-
l
i
n
g
u
a
l
t
r
a
n
sf
e
r
w
i
t
h
l
a
n
g
u
a
g
e
-
s
p
e
c
i
f
i
c
su
b
n
e
t
w
o
r
k
s
f
o
r
l
o
w
-
r
e
s
o
u
r
c
e
d
e
p
e
n
d
e
n
c
y
p
a
r
si
n
g
,
”
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
9
,
n
o
.
3
,
p
p
.
6
1
3
–
6
4
1
,
S
e
p
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
6
2
/
c
o
l
i
_
a
_
0
0
4
8
2
.
[
9
]
Z.
H
u
a
n
g
,
P
.
Y
u
,
a
n
d
J.
A
l
l
a
n
,
“
I
mp
r
o
v
i
n
g
c
r
o
ss
-
l
i
n
g
u
a
l
i
n
f
o
r
m
a
t
i
o
n
r
e
t
r
i
e
v
a
l
o
n
l
o
w
-
r
e
so
u
r
c
e
l
a
n
g
u
a
g
e
s
v
i
a
o
p
t
i
m
a
l
t
r
a
n
s
p
o
r
t
d
i
s
t
i
l
l
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
S
i
x
t
e
e
n
t
h
A
C
M
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
We
b
S
e
a
rc
h
a
n
d
D
a
t
a
M
i
n
i
n
g
,
F
e
b
.
2
0
2
3
,
p
p
.
1
0
4
8
–
1
0
5
6
,
d
o
i
:
1
0
.
1
1
4
5
/
3
5
3
9
5
9
7
.
3
5
7
0
4
6
8
.
[
1
0
]
S
.
M
a
n
a
f
i
a
n
d
N
.
K
r
i
s
h
n
a
sw
a
m
y
,
“
C
r
o
ss
-
l
i
n
g
u
a
l
t
r
a
n
sf
e
r
r
o
b
u
s
t
n
e
ss
t
o
l
o
w
e
r
-
r
e
so
u
r
c
e
l
a
n
g
u
a
g
e
s
o
n
a
d
v
e
r
sari
a
l
d
a
t
a
se
t
s,”
a
r
Xi
v
p
re
p
ri
n
t
a
rXi
v
:
2
4
0
3
.
2
0
0
5
6
,
2
0
2
4
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
4
0
3
.
2
0
0
5
6
.
[
1
1
]
H
.
Th
a
n
g
a
r
a
j
,
A
.
C
h
e
n
a
t
,
J.
S
.
W
a
l
i
a
,
a
n
d
V
.
M
a
r
i
v
a
t
e
,
“
C
r
o
s
s
-
l
i
n
g
u
a
l
t
r
a
n
sf
e
r
o
f
m
u
l
t
i
l
i
n
g
u
a
l
m
o
d
e
l
s
o
n
l
o
w
r
e
so
u
r
c
e
A
f
r
i
c
a
n
La
n
g
u
a
g
e
s
,
”
a
rX
i
v
p
r
e
p
r
i
n
t
a
rX
i
v
:
2
4
0
9
.
1
0
9
6
5
,
2
0
2
4
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
4
0
9
.
1
0
9
6
5
.
[
1
2
]
S
.
R
u
d
e
r
,
I
.
V
u
l
i
ć
,
a
n
d
A
.
S
ø
g
a
a
r
d
,
“
A
su
r
v
e
y
o
f
c
r
o
ss
-
l
i
n
g
u
a
l
w
o
r
d
e
mb
e
d
d
i
n
g
mo
d
e
l
s,
”
J
o
u
r
n
a
l
o
f
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
Re
se
a
rc
h
,
v
o
l
.
6
5
,
p
p
.
5
6
9
–
6
3
1
,
A
u
g
.
2
0
1
9
,
d
o
i
:
1
0
.
1
6
1
3
/
j
a
i
r
.
1
.
1
1
6
4
0
.
[
1
3
]
A
.
R
.
Jafar
i
,
B
.
H
e
i
d
a
r
y
,
R
.
F
a
r
a
h
b
a
k
h
s
h
,
M
.
S
a
l
e
h
i
,
a
n
d
M
.
Ja
l
i
l
i
,
“
Tr
a
n
sf
e
r
l
e
a
r
n
i
n
g
f
o
r
m
u
l
t
i
-
l
i
n
g
u
a
l
t
a
s
k
s
--
a
su
r
v
e
y
,
”
a
rX
i
v
p
re
p
ri
n
t
a
rXi
v
:
2
1
1
0
.
0
2
0
5
2
,
2
0
2
1
.
[
1
4
]
F
.
B
i
a
n
c
h
i
,
S
.
Te
r
r
a
g
n
i
,
D
.
H
o
v
y
,
D
.
N
o
z
z
a
,
a
n
d
E.
F
e
r
si
n
i
,
“
C
r
o
ss
-
l
i
n
g
u
a
l
c
o
n
t
e
x
t
u
a
l
i
z
e
d
t
o
p
i
c
m
o
d
e
l
s
w
i
t
h
z
e
r
o
-
sh
o
t
l
e
a
r
n
i
n
g
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
1
6
t
h
C
o
n
f
e
re
n
c
e
o
f
t
h
e
Eu
r
o
p
e
a
n
C
h
a
p
t
e
r
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s:
M
a
i
n
V
o
l
u
m
e
,
2
0
2
1
,
p
p
.
1
6
7
6
–
1
6
8
3
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
1
.
e
a
c
l
-
ma
i
n
.
1
4
3
.
[
1
5
]
R
.
H
u
i
d
r
o
m
a
n
d
Y
.
Le
p
a
g
e
,
“
Z
e
r
o
-
sh
o
t
t
r
a
n
sl
a
t
i
o
n
a
m
o
n
g
I
n
d
i
a
n
l
a
n
g
u
a
g
e
s,”
i
n
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
3
rd
W
o
rks
h
o
p
o
n
T
e
c
h
n
o
l
o
g
i
e
s
f
o
r
MT
o
f
L
o
w
R
e
so
u
rc
e
L
a
n
g
u
a
g
e
s
,
2
0
2
0
,
p
p
.
4
7
–
5
4
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
l
o
r
e
sm
t
-
1
.
7
.
[
1
6
]
D
.
K
a
k
w
a
n
i
e
t
a
l
.
,
“
I
n
d
i
c
N
LPS
u
i
t
e
:
mo
n
o
l
i
n
g
u
a
l
c
o
r
p
o
r
a
,
e
v
a
l
u
a
t
i
o
n
b
e
n
c
h
m
a
r
k
s
a
n
d
p
r
e
-
t
r
a
i
n
e
d
m
u
l
t
i
l
i
n
g
u
a
l
l
a
n
g
u
a
g
e
m
o
d
e
l
s
f
o
r
I
n
d
i
a
n
l
a
n
g
u
a
g
e
s
,
”
i
n
Fi
n
d
i
n
g
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:
EM
N
L
P
2
0
2
0
,
2
0
2
0
,
p
p
.
4
9
4
8
–
4
9
6
1
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
f
i
n
d
i
n
g
s
-
e
m
n
l
p
.
4
4
5
.
[
1
7
]
F
.
F
e
n
g
,
Y
.
Y
a
n
g
,
D
.
C
e
r
,
N
.
A
r
i
v
a
z
h
a
g
a
n
,
a
n
d
W
.
W
a
n
g
,
“
La
n
g
u
a
g
e
-
a
g
n
o
st
i
c
B
E
R
T
s
e
n
t
e
n
c
e
e
m
b
e
d
d
i
n
g
,
”
i
n
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
6
0
t
h
A
n
n
u
a
l
Me
e
t
i
n
g
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
(
Vo
l
u
m
e
1
:
L
o
n
g
P
a
p
e
rs)
,
2
0
2
2
,
p
p
.
8
7
8
–
8
9
1
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
2
.
a
c
l
-
l
o
n
g
.
6
2
.
[
1
8
]
M
.
F
a
r
o
u
k
,
“
M
e
a
su
r
i
n
g
se
n
t
e
n
c
e
s
s
i
mi
l
a
r
i
t
y
:
a
s
u
r
v
e
y
,
”
I
n
d
i
a
n
J
o
u
rn
a
l
o
f
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
1
2
,
n
o
.
2
5
,
p
p
.
1
–
1
1
,
J
u
l
.
2
0
1
9
,
d
o
i
:
1
0
.
1
7
4
8
5
/
i
j
s
t
/
2
0
1
9
/
v
1
2
i
2
5
/
1
4
3
9
7
7
.
[
1
9
]
A
.
C
o
n
n
e
a
u
e
t
a
l
.
,
“
U
n
s
u
p
e
r
v
i
s
e
d
c
r
o
ss
-
l
i
n
g
u
a
l
r
e
p
r
e
se
n
t
a
t
i
o
n
l
e
a
r
n
i
n
g
a
t
s
c
a
l
e
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s o
f
t
h
e
5
8
t
h
An
n
u
a
l
M
e
e
t
i
n
g
o
f
t
h
e
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
,
2
0
2
0
,
p
p
.
8
4
4
0
–
8
4
5
1
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
a
c
l
-
mai
n
.
7
4
7
.
[
2
0
]
A
.
C
o
n
n
e
a
u
e
t
a
l
.
,
“
X
N
LI
:
e
v
a
l
u
a
t
i
n
g
c
r
o
ss
-
l
i
n
g
u
a
l
se
n
t
e
n
c
e
r
e
p
r
e
se
n
t
a
t
i
o
n
s,”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
1
8
C
o
n
f
e
r
e
n
c
e
o
n
Em
p
i
r
i
c
a
l
M
e
t
h
o
d
s
i
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pr
o
c
e
ssi
n
g
,
2
0
1
8
,
p
p
.
2
4
7
5
–
2
4
8
5
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
D
1
8
-
1
2
6
9
.
[
2
1
]
D
.
A
g
g
a
r
w
a
l
,
V
.
G
u
p
t
a
,
a
n
d
A
.
K
u
n
c
h
u
k
u
t
t
a
n
,
“
I
n
d
i
c
X
N
LI
:
e
v
a
l
u
a
t
i
n
g
m
u
l
t
i
l
i
n
g
u
a
l
i
n
f
e
r
e
n
c
e
f
o
r
I
n
d
i
a
n
l
a
n
g
u
a
g
e
s,”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
2
2
C
o
n
f
e
re
n
c
e
o
n
Em
p
i
r
i
c
a
l
Me
t
h
o
d
s
i
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pr
o
c
e
ssi
n
g
,
2
0
2
2
,
p
p
.
1
0
9
9
4
–
1
1
0
0
6
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
2
.
e
mn
l
p
-
mai
n
.
7
5
5
.
[
2
2
]
L.
D
i
n
g
,
L
.
W
a
n
g
,
a
n
d
D
.
Ta
o
,
“
S
e
l
f
-
a
t
t
e
n
t
i
o
n
w
i
t
h
c
r
o
ss
-
l
i
n
g
u
a
l
p
o
s
i
t
i
o
n
r
e
p
r
e
se
n
t
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
5
8
t
h
An
n
u
a
l
Me
e
t
i
n
g
o
f
t
h
e
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
,
2
0
2
0
,
p
p
.
1
6
7
9
–
1
6
8
5
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
a
c
l
-
ma
i
n
.
1
5
3
.
[
2
3
]
L.
D
i
n
g
,
D
.
W
u
,
a
n
d
D
.
T
a
o
,
“
I
mp
r
o
v
i
n
g
n
e
u
r
a
l
m
a
c
h
i
n
e
t
r
a
n
sl
a
t
i
o
n
b
y
b
i
d
i
r
e
c
t
i
o
n
a
l
t
r
a
i
n
i
n
g
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
2
1
C
o
n
f
e
re
n
c
e
o
n
Em
p
i
ri
c
a
l
Me
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
,
2
0
2
1
,
p
p
.
3
2
7
8
–
3
2
8
4
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
1
.
e
mn
l
p
-
mai
n
.
2
6
3
.
[
2
4
]
L.
D
i
n
g
,
L
.
W
a
n
g
,
X
.
L
i
u
,
D
.
F
.
W
o
n
g
,
D
.
Ta
o
,
a
n
d
Z.
T
u
,
“
U
n
d
e
r
st
a
n
d
i
n
g
a
n
d
i
m
p
r
o
v
i
n
g
l
e
x
i
c
a
l
c
h
o
i
c
e
i
n
n
o
n
-
a
u
t
o
r
e
g
r
e
s
si
v
e
t
r
a
n
s
l
a
t
i
o
n
,
”
a
r
Xi
v
p
re
p
ri
n
t
a
r
Xi
v
:
2
0
1
2
.
1
4
5
8
3
,
2
0
2
0
.
[
2
5
]
L.
D
i
n
g
,
L.
W
a
n
g
,
D
.
W
u
,
D
.
Ta
o
,
a
n
d
Z.
T
u
,
“
C
o
n
t
e
x
t
-
a
w
a
r
e
c
r
o
ss
-
a
t
t
e
n
t
i
o
n
f
o
r
n
o
n
-
a
u
t
o
r
e
g
r
e
ssi
v
e
t
r
a
n
sl
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
2
8
t
h
I
n
t
e
rn
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
,
2
0
2
0
,
p
p
.
4
3
9
6
–
4
4
0
2
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
c
o
l
i
n
g
-
mai
n
.
3
8
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
2
,
Ap
r
il
20
2
6
:
9
7
3
-
980
980
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Nir
a
n
ja
n
G
C
is
a
p
o
st
-
g
ra
d
u
a
te
stu
d
e
n
t
a
t
Vis
v
e
sv
a
ra
y
a
Tec
h
n
o
l
o
g
ica
l
Un
iv
e
rsity
(VTU),
p
u
rsu
i
n
g
M
.
Tec
h
in
c
o
m
p
u
ter
sc
ien
c
e
a
n
d
e
n
g
in
e
e
ri
n
g
a
t
Ra
sh
tree
y
a
Vid
y
a
lay
a
C
o
ll
e
g
e
o
f
E
n
g
in
e
e
rin
g
(RVCE)
Ba
n
g
a
lo
re
,
In
d
ia.
Wi
t
h
a
stro
n
g
i
n
c
li
n
a
ti
o
n
to
wa
rd
s
g
e
n
e
ra
ti
v
e
AI
a
n
d
larg
e
lan
g
u
a
g
e
m
o
d
e
ls
(LL
M
)s,
h
is
a
re
a
s
o
f
i
n
tere
st
in
c
lu
d
e
DL,
NLP
,
a
n
d
trad
i
ti
o
n
a
l
m
a
c
h
in
e
lea
rn
in
g
(M
L).
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
n
iran
ja
n
g
c
.
sc
s2
3
@r
v
c
e
.
e
d
u
.
in
.
Ra
m
a
k
a
n
t
h
K
u
m
a
r
P
is
c
u
rre
n
tl
y
a
p
r
o
fe
ss
o
r
a
n
d
t
h
e
d
e
a
n
o
f
th
e
CS
E
Cl
u
ste
r,
Ra
sh
tree
y
a
Vid
y
a
lay
a
Co
ll
e
g
e
o
f
En
g
in
e
e
rin
g
(RVCE),
Ba
n
g
a
lo
re
,
In
d
ia.
He
h
a
s
tau
g
h
t
c
o
u
rse
s
o
n
n
e
two
r
k
p
ro
g
ra
m
m
in
g
a
n
d
c
y
b
e
rse
c
u
rit
y
f
o
r
In
d
u
stry
4
.
0
.
He
h
a
s
p
u
b
li
s
h
e
d
m
o
re
t
h
a
n
1
0
0
re
se
a
rc
h
a
rti
c
les
.
He
is
a
se
n
io
r
m
e
m
b
e
r
a
t
IEE
E
a
n
d
h
a
s
e
x
e
c
u
te
d
se
v
e
ra
l
f
u
n
d
e
d
re
se
a
rc
h
a
n
d
c
o
n
su
lt
a
n
c
y
p
r
o
jec
ts
sp
o
n
so
re
d
b
y
DRD
O,
IS
RO,
CAIR,
LRDE
,
AICTE,
G
E
In
d
ia
P
v
t
.
Lt
d
.
,
CABS,
a
n
d
HPE
.
His
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
d
i
g
it
a
l
ima
g
e
p
r
o
c
e
ss
in
g
,
p
a
tt
e
r
n
re
c
o
g
n
it
io
n
,
a
n
d
n
a
t
u
ra
l
lan
g
u
a
g
e
p
ro
c
e
ss
in
g
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
ra
m
a
k
a
n
th
k
p
@r
v
c
e
.
e
d
u
.
i
n
.
Pa
v
it
h
r
a
H
is
c
u
rre
n
tl
y
a
n
a
ss
o
c
iate
p
ro
fe
ss
o
r
i
n
t
h
e
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
En
g
i
n
e
e
rin
g
De
p
a
rtme
n
t
a
t
Ra
sh
tree
y
a
Vid
y
a
lay
a
C
o
ll
e
g
e
o
f
En
g
in
e
e
rin
g
(RVCE),
Ba
n
g
a
lo
re
,
In
d
ia.
He
r
re
se
a
rc
h
i
n
tere
sts
a
re
so
ftwa
re
d
e
fin
e
d
n
e
tw
o
rk
s,
m
a
c
h
in
e
lea
rn
in
g
,
d
e
e
p
lea
rn
in
g
,
so
ftwa
re
e
n
g
in
e
e
rin
g
.
S
h
e
h
a
s
e
x
e
c
u
ted
p
r
o
jec
ts
sp
o
n
so
re
d
b
y
S
a
m
su
n
g
,
T
o
y
o
ta.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
p
a
v
it
h
ra
h
@rv
c
e
.
e
d
u
.
in
.
Mi
n
a
l
Mo
h
a
r
ir
is
c
u
rre
n
tl
y
a
p
ro
fe
ss
o
r
i
n
th
e
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter S
c
ien
c
e
a
n
d
En
g
i
n
e
e
rin
g
a
t
Ra
sh
tree
y
a
Vi
d
y
a
lay
a
Co
l
leg
e
o
f
E
n
g
in
e
e
rin
g
(RV
CE),
Be
n
g
a
lu
r
u
,
with
o
v
e
r
1
4
y
e
a
rs
o
f
a
c
a
d
e
m
ic
e
x
p
e
rien
c
e
.
S
h
e
h
o
l
d
s
a
P
h
.
D.
in
c
o
m
p
u
ter
sc
ien
c
e
a
n
d
e
n
g
i
n
e
e
rin
g
a
n
d
a
n
M
.
T
e
c
h
in
c
o
m
p
u
ter
n
e
two
rk
e
n
g
in
e
e
rin
g
.
He
r
a
re
a
s
o
f
e
x
p
e
rti
se
in
c
lu
d
e
c
o
m
p
u
ter
n
e
two
rk
s,
c
y
b
e
rse
c
u
rit
y
,
a
n
d
h
ig
h
-
p
e
rf
o
rm
a
n
c
e
c
o
m
p
u
ti
n
g
.
S
h
e
h
a
s
p
u
b
li
s
h
e
d
e
x
ten
si
v
e
ly
,
with
o
v
e
r
7
0
p
a
p
e
rs
i
n
re
p
u
ted
i
n
tern
a
ti
o
n
a
l
a
n
d
n
a
ti
o
n
a
l
j
o
u
r
n
a
ls
a
n
d
c
o
n
fe
re
n
c
e
s,
a
n
d
h
a
s
g
u
id
e
d
n
u
m
e
ro
u
s
UG
a
n
d
P
G
re
se
a
r
c
h
p
ro
jec
ts.
S
h
e
h
a
s
led
se
v
e
ra
l
fu
n
d
e
d
re
se
a
rc
h
a
n
d
c
o
n
su
lt
a
n
c
y
p
r
o
jec
ts
in
c
o
ll
a
b
o
ra
ti
o
n
with
o
rg
a
n
iza
ti
o
n
s l
ik
e
NV
IDIA
,
Cit
rix
,
S
a
m
su
n
g
,
a
n
d
DRD
O l
a
b
s,
se
c
u
ri
n
g
g
ra
n
ts
wo
rth
se
v
e
ra
l
lak
h
s.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
i
l:
m
i
n
a
lmo
h
a
rir
@rv
c
e
.
e
d
u
.
i
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.