I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
2
,
A
p
r
il 2
0
2
6
,
p
p
.
1
9
0
9
~
1
9
1
8
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
2
.
p
p
1
9
0
9
-
1
9
1
8
1909
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Ev
a
lua
ting do
cument c
hunking
a
ppro
a
ches for r
et
riev
a
l
-
a
ug
mented gener
a
tion in edi
toria
l
co
ntent
E
rwa
nn
L
a
v
a
re
c,
Yu D
u
D
e
p
a
r
t
me
n
t
o
f
R
e
s
e
a
r
c
h
a
n
d
D
e
v
e
l
o
p
men
t
,
C
l
o
u
d
i
s
M
i
n
e
,
M
o
n
t
p
e
l
l
i
e
r
,
F
r
a
n
c
e
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Sep
2
4
,
2
0
2
5
R
ev
is
ed
J
an
1
3
,
2
0
2
6
Acc
ep
ted
Feb
6
,
2
0
2
6
Re
tri
e
v
a
l
-
a
u
g
m
e
n
ted
g
e
n
e
ra
ti
o
n
(
RAG
)
sy
ste
m
s
p
ro
m
ise
g
ro
u
n
d
e
d
a
n
sw
e
rs
fro
m
larg
e
lan
g
u
a
g
e
m
o
d
e
ls
(LL
M
s),
y
e
t
p
e
rfo
rm
a
n
c
e
d
e
p
e
n
d
s
c
rit
ica
ll
y
o
n
h
o
w
so
u
rc
e
d
o
c
u
m
e
n
ts
a
re
s
e
g
m
e
n
ted
b
e
fo
re
in
d
e
x
i
n
g
.
T
h
is
stu
d
y
in
v
e
stig
a
tes
h
o
w
p
re
-
i
n
d
e
x
c
h
u
n
k
in
g
stra
teg
ies
a
ffe
c
t
b
o
t
h
re
tri
e
v
a
l
a
c
c
u
ra
c
y
a
n
d
a
n
sw
e
r
q
u
a
li
t
y
in
d
o
m
a
in
-
sp
e
c
ifi
c
sc
e
n
a
rio
s.
We
c
u
ra
ted
a
c
o
rp
u
s
o
n
so
ftwa
re
-
as
-
a
-
se
rv
ice
(S
a
a
S
)
e
d
it
o
rial
c
o
n
ten
t
a
n
d
c
o
n
stru
c
ted
a
h
i
g
h
-
q
u
a
li
t
y
e
v
a
lu
a
ti
o
n
d
a
tas
e
t
c
o
n
tai
n
in
g
2
,
4
1
9
q
u
e
stio
n
-
a
n
sw
e
r
(QA
)
p
a
irs
g
e
n
e
ra
ted
th
ro
u
g
h
a
u
to
m
a
ted
p
r
o
m
p
ti
n
g
a
n
d
q
u
a
l
it
y
c
o
n
tro
l.
We
c
o
m
p
a
re
d
fo
u
r
c
h
u
n
k
in
g
a
p
p
ro
a
c
h
e
s
,
in
c
lu
d
i
n
g
fix
e
d
-
siz
e
,
stru
c
t
u
re
-
a
wa
re
re
c
u
rsiv
e
,
se
m
a
n
ti
c
,
a
n
d
LL
M
-
b
a
se
d
m
e
th
o
d
s
.
Ou
r
e
v
a
l
u
a
ti
o
n
p
ro
t
o
c
o
l
a
ss
e
ss
e
d
re
tri
e
v
a
l
th
ro
u
g
h
d
o
c
u
m
e
n
t
lo
c
a
li
z
a
ti
o
n
,
se
m
a
n
ti
c
sim
il
a
rit
y
,
a
n
d
c
o
n
tex
t
re
lev
a
n
c
e
,
wh
il
e
g
e
n
e
ra
ti
o
n
q
u
a
li
ty
wa
s
e
v
a
lu
a
ted
u
si
n
g
c
h
a
in
-
of
-
th
o
u
g
h
t
(Co
T)
c
rit
e
ria
d
riv
e
n
b
y
j
u
d
g
m
e
n
ts
fro
m
LL
M
s.
Re
su
lt
s
d
e
m
o
n
stra
te
th
a
t
re
c
u
rsiv
e
c
h
u
n
k
i
n
g
c
o
n
siste
n
tl
y
o
u
tp
e
rfo
rm
s
o
t
h
e
r
a
p
p
r
o
a
c
h
e
s
a
c
ro
ss
a
ll
m
e
tri
c
s.
S
m
a
ll
e
r
c
h
u
n
k
s
imp
ro
v
e
d
o
c
u
m
e
n
t
l
o
c
a
li
z
a
ti
o
n
,
wh
il
e
m
o
d
e
ra
tely
larg
e
r
c
h
u
n
k
s
e
n
h
a
n
c
e
se
m
a
n
ti
c
a
li
g
n
m
e
n
t
a
n
d
g
e
n
e
ra
ti
o
n
sc
o
r
e
s.
LL
M
-
b
a
se
d
c
h
u
n
k
in
g
v
a
rian
ts
sh
o
w
c
o
m
p
e
ti
ti
v
e
p
e
rfo
rm
a
n
c
e
b
u
t
d
o
n
o
t
e
x
c
e
e
d
to
p
re
c
u
rsiv
e
c
o
n
fig
u
ra
ti
o
n
s
o
n
th
e
d
a
tas
e
t.
Th
e
se
fin
d
in
g
s
in
d
ica
te
th
a
t
p
re
se
rv
in
g
d
o
c
u
m
e
n
t
str
u
c
tu
re
t
h
ro
u
g
h
re
c
u
rsi
v
e
c
h
u
n
k
in
g
is
b
e
n
e
ficia
l
fo
r
p
ra
c
ti
c
a
l
RAG
imp
lem
e
n
tatio
n
s,
p
r
o
v
i
d
i
n
g
a
c
ti
o
n
a
b
le
g
u
i
d
a
n
c
e
fo
r
c
h
u
n
k
siz
e
se
lec
ti
o
n
w
h
il
e
h
i
g
h
li
g
h
t
in
g
to
k
e
n
-
b
u
d
g
e
t
c
o
n
stra
i
n
ts
in
c
u
r
re
n
t
l
o
n
g
-
c
o
n
tex
t
m
o
d
e
ls.
K
ey
w
o
r
d
s
:
Do
cu
m
en
t c
h
u
n
k
i
n
g
I
n
f
o
r
m
atio
n
r
etr
ie
v
al
L
ar
g
e
lan
g
u
ag
e
m
o
d
el
Natu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
Qu
esti
o
n
an
s
wer
in
g
R
etr
iev
al
-
au
g
m
en
ted
g
en
er
ati
o
n
T
ex
t seg
m
en
tatio
n
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
:
Yu
Du
Dep
ar
tm
en
t o
f
R
esear
ch
an
d
Dev
elo
p
m
en
t,
C
lo
u
d
is
Min
e
C
A
C
en
ter
,
6
2
1
R
u
e
Geo
r
g
es
Mé
liès
,
3
4
0
0
0
M
o
n
tp
ellier
,
Fr
an
ce
E
m
ail:
y
u
.
d
u
@
ap
p
v
izer
.
co
m
1.
I
NT
RO
D
UCT
I
O
N
W
h
ile
lar
g
e
lan
g
u
a
g
e
m
o
d
els
(
L
L
Ms)
h
av
e
tr
an
s
f
o
r
m
ed
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
ca
p
ab
ilit
ies,
th
ey
r
em
ain
v
u
ln
er
ab
le
to
h
a
llu
cin
atio
n
an
d
f
ac
tu
al
d
r
if
t
wh
en
q
u
er
ies
ex
ten
d
b
ey
o
n
d
th
eir
tr
ain
in
g
d
ata
b
o
u
n
d
ar
ies
[
1
]
.
R
etr
iev
al
-
au
g
m
en
ted
g
en
er
atio
n
(
R
AG)
e
m
er
g
ed
as
a
c
r
itical
s
o
lu
tio
n
to
t
h
is
lim
itatio
n
,
s
y
s
tem
atica
lly
g
r
o
u
n
d
i
n
g
L
L
M
r
esp
o
n
s
es
in
ex
te
r
n
al
ev
i
d
e
n
ce
b
y
in
teg
r
atin
g
d
e
d
icate
d
r
etr
iev
al
m
ec
h
an
is
m
s
with
g
en
er
ativ
e
m
o
d
els
[
2
]
–
[
4
]
.
Ho
wev
e
r
,
r
ea
l
-
wo
r
ld
d
e
p
l
o
y
m
en
ts
o
f
R
AGs
f
ac
e
a
f
u
n
d
am
en
tal
ch
allen
g
e:
th
ey
m
u
s
t
p
r
o
ce
s
s
ex
ten
s
iv
e,
s
tr
u
ctu
r
ally
co
m
p
lex
d
o
c
u
m
en
ts
co
n
tain
in
g
d
iv
er
s
e
elem
en
ts
,
in
clu
d
in
g
h
ea
d
in
g
s
,
p
a
r
ag
r
a
p
h
s
,
lis
ts
,
an
d
tab
les.
I
n
th
ese
p
r
ac
tical
ap
p
licatio
n
s
,
s
y
s
tem
p
er
f
o
r
m
a
n
c
e
d
ep
en
d
s
n
o
t
o
n
l
y
o
n
r
etr
ie
v
er
s
o
p
h
is
ticatio
n
an
d
L
L
M
ca
p
ab
ilit
ies,
b
u
t
also
o
n
t
h
e
o
f
ten
-
o
v
er
lo
o
k
ed
p
r
o
ce
s
s
o
f
d
o
cu
m
en
t
ch
u
n
k
i
n
g
d
u
r
in
g
p
r
e
-
in
d
ex
in
g
.
T
h
is
ch
u
n
k
in
g
d
ir
ec
tly
d
e
ter
m
in
es
wh
at
co
n
ten
t
b
ec
o
m
es
av
ailab
le
to
th
e
r
etr
iev
er
a
n
d
c
o
n
s
eq
u
e
n
tly
,
wh
at
co
n
tex
t
r
ea
ch
es
th
e
L
L
M,
m
ak
in
g
it
a
p
iv
o
tal
f
ac
t
o
r
in
b
o
th
r
etr
iev
al
r
elev
an
ce
an
d
f
in
al
a
n
s
wer
q
u
ality
.
B
y
an
alo
g
y
with
elec
tr
o
n
ic
s
y
s
tem
s
,
wh
er
e
th
e
f
ir
s
t
am
p
lific
atio
n
s
tag
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
9
0
9
-
1
9
1
8
1910
lar
g
ely
d
eter
m
in
es
th
e
s
ig
n
al
q
u
ality
th
r
o
u
g
h
o
u
t
th
e
p
r
o
ce
s
s
in
g
ch
ain
,
th
e
a
u
th
o
r
s
p
o
s
it
th
at
th
e
ef
f
ec
tiv
en
ess
an
d
r
o
b
u
s
tn
ess
o
f
t
h
e
in
itial
p
r
o
ce
s
s
in
g
s
tag
e
in
a
R
AG
p
ip
elin
e
p
lay
a
d
ec
is
iv
e
r
o
le
in
th
e
o
v
er
all
q
u
ality
o
f
th
e
g
en
er
ated
r
esp
o
n
s
es.
Desp
ite
it
s
f
u
n
d
am
en
tal
im
p
o
r
tan
ce
,
d
o
cu
m
e
n
t
ch
u
n
k
in
g
r
em
ain
s
p
o
o
r
ly
u
n
d
er
s
to
o
d
an
d
in
ad
eq
u
ately
v
alid
ated
ac
r
o
s
s
d
o
m
ain
-
s
p
ec
if
ic
a
p
p
licatio
n
s
.
Pra
ctitio
n
er
s
ty
p
ically
r
eso
r
t t
o
g
en
er
ic
h
eu
r
is
tics
,
e.
g
.
,
“u
s
e
1
K
ch
ar
ac
ter
s
with
5
-
1
0
%
o
v
e
r
lap
”,
wh
ile
r
esea
r
ch
ef
f
o
r
ts
co
n
ce
n
tr
ate
p
r
im
ar
ily
o
n
d
o
wn
s
tr
ea
m
co
m
p
o
n
en
ts
lik
e
r
etr
iev
al
alg
o
r
ith
m
s
,
r
er
an
k
in
g
m
ec
h
an
is
m
s
[
5
]
,
an
d
g
e
n
er
ato
r
p
r
o
m
p
t
in
g
s
tr
ateg
ies.
T
h
is
is
s
u
e
is
p
ar
ticu
lar
ly
p
r
o
b
lem
atic
g
iv
en
th
at
ch
u
n
k
in
g
d
ec
is
io
n
s
em
b
o
d
y
a
f
u
n
d
am
e
n
tal
tr
ad
e
-
o
f
f
with
n
o
u
n
iv
er
s
al
s
o
lu
tio
n
:
s
m
aller
ch
u
n
k
s
en
h
an
ce
p
r
ec
is
e
lo
ca
li
za
tio
n
with
in
a
d
o
cu
m
en
t
b
u
t
r
is
k
f
r
ag
m
en
tin
g
co
h
er
en
t
m
ea
n
i
n
g
ac
r
o
s
s
ar
tific
ial
b
o
u
n
d
ar
ies;
lar
g
er
c
h
u
n
k
s
p
r
eser
v
e
to
p
ical
co
n
t
ex
t
an
d
s
em
an
tic
r
elatio
n
s
h
ip
s
b
u
t
i
n
tr
o
d
u
ce
ir
r
elev
an
t
n
o
is
e
an
d
s
tr
ain
to
k
en
b
u
d
g
ets,
co
m
p
o
u
n
d
in
g
th
e
l
o
n
g
-
c
o
n
tex
t
r
eliab
ilit
y
ch
allen
g
es
th
at
p
la
g
u
e
m
o
d
er
n
L
L
Ms
[
6
]
.
T
h
e
c
o
m
p
lex
ity
d
ee
p
en
s
wh
e
n
co
n
s
id
er
in
g
th
e
d
iv
er
s
e
m
eth
o
d
o
l
o
g
ical
ap
p
r
o
ac
h
es
av
ailab
le.
Fix
ed
-
s
ize
win
d
o
ws,
s
tr
u
ctu
r
e
-
awa
r
e
r
ec
u
r
s
iv
e
s
eg
m
en
tatio
n
,
s
em
an
tic
b
o
u
n
d
ar
y
d
etec
tio
n
,
an
d
L
L
M
-
d
r
iv
en
s
eg
m
e
n
tatio
n
ea
ch
in
v
o
lv
e
d
is
tin
ct
tr
ad
e
-
o
f
f
s
,
wh
ich
m
ay
in
ter
ac
t
with
d
o
m
ain
ch
ar
ac
ter
is
tics
an
d
d
o
c
u
m
en
t
ty
p
es
i
n
u
n
p
r
e
d
ictab
le
way
s
.
Yet
co
m
p
r
e
h
en
s
iv
e,
e
n
d
-
to
-
en
d
ev
alu
atio
n
s
th
at
s
y
s
tem
atica
lly
co
n
n
ec
t
ch
u
n
k
in
g
s
tr
ateg
ies
to
b
o
th
r
etr
iev
al
ef
f
ec
tiv
en
ess
an
d
an
s
wer
q
u
ality
with
in
s
p
ec
if
ic
d
o
m
ain
s
r
e
m
ain
co
n
s
p
ic
u
o
u
s
ly
a
b
s
en
t
f
r
o
m
t
h
e
liter
atu
r
e.
W
h
ile
r
ec
en
t
s
u
r
v
ey
s
a
n
d
b
en
ch
m
ar
k
s
h
av
e
e
x
p
an
d
ed
R
AG
ev
alu
atio
n
f
r
am
ewo
r
k
s
[
7
]
–
[
9
]
,
ac
tio
n
ab
le
g
u
id
a
n
c
e
f
o
r
p
r
ac
titi
o
n
er
s
wo
r
k
in
g
with
s
p
ec
ialized
p
r
o
f
ess
io
n
al
co
r
p
o
r
a
r
em
ain
s
elu
s
iv
e.
T
h
is
w
o
r
k
ad
d
r
ess
es
t
h
is
c
r
i
tic
al
g
a
p
t
h
r
o
u
g
h
a
co
m
p
r
e
h
e
n
s
i
v
e
ev
al
u
a
ti
o
n
in
t
h
e
ed
it
o
r
ial
d
o
m
ai
n
t
h
a
t
ad
d
r
ess
es
t
h
e
t
o
p
ic
o
f
s
o
f
twa
r
e
-
as
-
a
-
s
er
v
i
ce
(
SaaS
)
.
W
e
s
y
s
t
em
a
tic
all
y
c
u
r
at
e
a
h
i
g
h
-
q
u
a
lit
y
d
atas
et
f
r
o
m
tw
o
esta
b
lis
h
ed
m
ed
ia
s
o
u
r
c
es i
n
t
h
e
Sa
aS
f
ie
ld
a
n
d
co
n
s
t
r
u
ct
a
r
ig
o
r
o
u
s
q
u
est
io
n
-
a
n
s
w
er
(
QA
)
ev
al
u
at
io
n
s
et
u
s
in
g
a
s
t
ate
-
of
-
t
h
e
-
ar
t
L
L
M
t
o
g
en
e
r
at
e
f
a
ct
-
s
e
ek
in
g
q
u
esti
o
n
s
d
i
r
ec
tl
y
g
r
o
u
n
d
e
d
i
n
s
o
u
r
ce
p
ass
a
g
es
,
wit
h
a
u
t
o
m
ate
d
q
u
ali
ty
c
o
n
tr
o
ls
en
s
u
r
i
n
g
d
a
taset
in
te
g
r
it
y
.
O
u
r
e
v
a
lu
ati
o
n
f
r
a
m
ew
o
r
k
i
n
t
r
o
d
u
c
es
a
m
u
lti
-
d
im
e
n
s
i
o
n
al
ass
ess
m
e
n
t
p
r
o
t
o
co
l
t
h
at
ca
p
t
u
r
es
b
o
th
r
et
r
ie
v
a
l
e
f
f
ec
ti
v
e
n
e
s
s
an
d
g
e
n
e
r
ati
o
n
q
u
alit
y
.
W
e
m
ea
s
u
r
e
r
et
r
i
ev
al
p
e
r
f
o
r
m
a
n
c
e
a
c
r
o
s
s
t
h
r
ee
c
o
m
p
le
m
e
n
ta
r
y
m
e
tr
ics:
i
)
d
o
c
u
m
e
n
t
-
le
v
e
l
l
o
ca
li
za
ti
o
n
,
ii
)
s
em
a
n
t
ic
al
ig
n
m
e
n
t
,
an
d
i
ii)
p
r
a
g
m
at
ic
r
e
le
v
a
n
c
e.
T
o
ass
ess
e
n
d
-
to
-
e
n
d
s
y
s
te
m
p
e
r
f
o
r
m
an
ce
,
w
e
e
m
p
lo
y
L
L
M
-
j
u
d
g
e
d
[
1
0
]
,
ch
ai
n
-
of
-
th
o
u
g
h
t
(
C
o
T
)
e
v
al
u
ati
o
n
o
f
f
i
n
al
a
n
s
we
r
q
u
ali
ty
,
p
r
o
v
i
d
i
n
g
in
s
ig
h
t
i
n
t
o
h
o
w
ch
u
n
k
in
g
d
e
cisi
o
n
s
u
lt
im
ate
ly
im
p
ac
t
u
s
e
r
-
f
a
ci
n
g
o
u
t
co
m
es
.
W
i
th
in
t
h
is
u
n
i
f
i
ed
e
x
p
e
r
im
e
n
ta
l
p
r
o
c
ess
,
we
co
n
d
u
ct
s
y
s
t
em
ati
c
co
m
p
a
r
is
o
n
s
ac
r
o
s
s
f
o
u
r
m
a
jo
r
c
h
u
n
k
i
n
g
f
am
ili
es,
i
n
cl
u
d
in
g
f
i
x
e
d
-
s
iz
e,
r
e
cu
r
s
i
v
e
(
s
t
r
u
ct
u
r
e
-
awa
r
e
)
,
s
em
a
n
ti
c
b
o
u
n
d
a
r
y
d
e
tec
ti
o
n
,
a
n
d
L
L
M
-
b
as
ed
a
p
p
r
o
ac
h
es
.
T
h
is
p
ap
er
d
eliv
er
s
th
r
ee
k
ey
co
n
tr
ib
u
tio
n
s
to
th
e
R
AG
ch
u
n
k
in
g
liter
atu
r
e:
a
m
eticu
lo
u
s
l
y
cu
r
ated
,
d
o
m
ain
-
s
p
ec
if
ic
QA
r
eso
u
r
ce
f
o
r
SaaS
e
d
ito
r
ial
co
n
ten
t
wit
h
r
ig
o
r
o
u
s
q
u
ality
co
n
tr
o
ls
;
a
h
o
lis
tic
ev
alu
atio
n
p
r
o
to
co
l
th
at
s
y
s
tem
atica
lly
co
n
n
ec
ts
ch
u
n
k
in
g
s
tr
ateg
ies
to
b
o
th
r
et
r
iev
al
ef
f
ec
tiv
en
e
s
s
an
d
f
in
al
an
s
wer
q
u
ality
;
an
d
ev
id
en
ce
-
b
ac
k
e
d
,
ac
tio
n
ab
le
g
u
id
an
ce
f
o
r
o
p
tim
izin
g
ch
u
n
k
s
ize
an
d
s
eg
m
en
t
atio
n
ap
p
r
o
ac
h
es
in
R
AG
s
y
s
tem
s
.
B
y
g
r
o
u
n
d
i
n
g
o
u
r
an
aly
s
is
in
a
r
ea
lis
tic
co
r
p
u
s
an
d
em
p
l
o
y
in
g
ev
al
u
atio
n
m
etr
ics
th
at
in
teg
r
ate
b
o
t
h
co
m
p
u
tatio
n
al
s
im
ilar
ity
m
ea
s
u
r
es
an
d
n
u
an
ce
d
L
L
M
ass
ess
m
en
ts
,
th
is
wo
r
k
b
r
id
g
es
th
e
g
a
p
b
etwe
en
th
eo
r
etica
l
c
h
u
n
k
in
g
r
esear
ch
an
d
p
r
ac
tical
d
ep
lo
y
m
en
t
n
ee
d
s
.
Ou
r
f
in
d
in
g
s
p
r
o
v
id
e
p
r
ac
titi
o
n
er
s
with
th
e
p
r
in
cip
led
,
em
p
ir
ically
v
alid
ated
d
esig
n
g
u
id
an
ce
n
ec
ess
ar
y
f
o
r
b
u
ild
in
g
R
AG
s
y
s
tem
s
th
at
ar
e
b
o
th
r
eliab
le
an
d
ef
f
icien
t in
r
ea
l
-
w
o
r
ld
ap
p
licatio
n
s
.
T
h
e
r
em
ain
d
e
r
o
f
th
e
p
ap
er
is
s
tr
u
ctu
r
ed
as
f
o
llo
ws.
Sect
io
n
2
r
ev
iews
p
r
io
r
w
o
r
k
an
d
o
u
tlin
es
th
e
f
am
ilies
o
f
ch
u
n
k
in
g
s
tr
at
eg
ies
co
n
s
id
er
ed
in
t
h
is
s
tu
d
y
.
Sectio
n
3
d
escr
ib
es
th
e
d
atas
et
co
n
s
tr
u
ctio
n
an
d
th
e
ev
alu
atio
n
p
r
o
t
o
co
l,
in
clu
d
in
g
r
etr
ie
v
al
an
d
g
en
er
atio
n
m
etr
ics.
Sectio
n
4
r
e
p
o
r
ts
th
e
em
p
ir
ical
r
esu
lts
an
d
d
is
cu
s
s
es
th
eir
im
p
licatio
n
s
an
d
tr
a
d
e
-
o
f
f
s
.
Sectio
n
5
c
o
n
clu
d
es
with
k
ey
ta
k
ea
way
s
an
d
d
ir
ec
tio
n
s
f
o
r
f
u
tu
r
e
r
esear
c
h
.
2.
RE
L
AT
E
D
WO
RK
S AN
D
CH
UNK
I
NG
ST
RAT
E
G
I
E
S
T
h
e
R
AG
ap
p
r
o
ac
h
in
teg
r
ates
an
in
f
o
r
m
atio
n
r
et
r
iev
al
co
m
p
o
n
en
t
with
an
L
L
M
to
g
en
er
ate
an
s
wer
s
th
at
ar
e
g
r
o
u
n
d
e
d
in
ex
ter
n
al
k
n
o
wled
g
e
[
2
]
,
[
3
]
.
As
illu
s
tr
ated
in
Fig
u
r
e
1
,
a
R
AG
s
y
s
tem
o
p
e
r
ates
th
r
o
u
g
h
th
r
ee
p
r
in
ci
p
al
p
h
ases
:
v
ec
to
r
d
atab
ase
b
u
ild
in
g
,
r
etr
ie
v
al,
a
n
d
g
e
n
er
atio
n
b
ased
o
n
L
L
M.
First,
in
th
e
v
ec
to
r
d
atab
ase
b
u
ild
in
g
p
h
ase,
a
tex
tu
al
co
r
p
u
s
is
s
eg
m
en
ted
in
to
s
m
aller
,
r
etr
iev
ab
le
u
n
its
u
s
in
g
a
ch
u
n
k
in
g
m
o
d
el.
E
ac
h
ch
u
n
k
is
th
en
tr
a
n
s
f
o
r
m
ed
in
to
a
d
e
n
s
e
n
u
m
er
ical
r
ep
r
esen
tatio
n
u
s
in
g
an
e
m
b
ed
d
in
g
m
o
d
el.
T
h
e
r
esu
ltin
g
em
b
e
d
d
in
g
s
ar
e
s
to
r
ed
in
a
v
ec
to
r
d
atab
ase
t
o
e
n
ab
le
ef
f
icien
t,
s
im
ilar
ity
-
b
as
ed
ac
ce
s
s
.
Seco
n
d
,
d
u
r
in
g
th
e
r
etr
iev
al
p
h
ase,
a
u
s
er
q
u
e
r
y
is
c
o
n
v
e
r
ted
i
n
to
i
ts
o
wn
em
b
e
d
d
in
g
r
e
p
r
esen
ta
tio
n
(
d
e
n
s
e
v
ec
to
r
)
u
s
in
g
th
e
s
am
e
em
b
ed
d
i
n
g
m
o
d
el,
a
n
d
a
s
em
an
tic
s
ea
r
ch
is
p
er
f
o
r
m
e
d
with
in
th
e
v
ec
to
r
d
atab
ase
to
id
en
tify
th
e
to
p
-
N
m
o
s
t
r
elev
an
t
ch
u
n
k
s
ac
co
r
d
in
g
to
co
s
in
e
s
im
ilar
ity
o
r
an
o
th
er
d
is
tan
ce
m
e
tr
ic.
T
o
ac
ce
ler
ate
lar
g
e
-
s
ca
le
r
etr
iev
al,
m
o
d
er
n
R
AG
s
y
s
tem
s
o
f
ten
r
ely
o
n
a
p
p
r
o
x
im
ate
n
ea
r
est
n
eig
h
b
o
r
(
ANN)
alg
o
r
ith
m
s
s
u
ch
as
h
ier
ar
ch
ical
n
av
ig
a
b
le
s
m
all
wo
r
ld
g
r
a
p
h
s
(
HNSW
)
[
1
1
]
,
i
n
v
er
ted
f
ile
in
d
e
x
es
(
I
VF)
[
1
2
]
,
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
va
lu
a
tin
g
d
o
cu
me
n
t c
h
u
n
kin
g
a
p
p
r
o
a
ch
es fo
r
r
etri
ev
a
l
-
a
u
g
men
ted
g
e
n
era
tio
n
in
…
(
E
r
w
a
n
n
La
va
r
ec
)
1911
lo
ca
lity
-
s
en
s
itiv
e
h
ash
i
n
g
(
L
SH)
[
1
3
]
.
T
h
ese
alg
o
r
ith
m
s
p
r
o
v
id
e
s
u
b
-
lin
ea
r
q
u
er
y
tim
e
wh
ile
m
ain
tain
in
g
h
ig
h
r
ec
all.
Fin
ally
,
in
th
e
g
e
n
er
atio
n
p
h
ase,
t
h
e
r
et
r
iev
ed
ch
u
n
k
s
a
r
e
c
o
n
ca
ten
ated
an
d
p
r
o
v
id
e
d
as
e
x
ter
n
al
co
n
tex
t
to
th
e
L
L
M,
wh
ich
s
y
n
th
esizes
a
f
in
al
an
s
wer
g
r
o
u
n
d
ed
in
th
e
r
etr
ie
v
ed
ev
id
e
n
c
e.
T
h
is
s
tep
en
s
u
r
es
f
ac
tu
al
co
n
s
is
ten
cy
an
d
r
ed
u
ce
s
h
allu
cin
atio
n
b
y
co
n
d
itio
n
in
g
th
e
m
o
d
el’
s
o
u
tp
u
t o
n
v
er
if
i
ed
co
n
ten
t.
Fig
u
r
e
1
.
Gen
e
r
al
wo
r
k
f
lo
w
o
f
a
R
AG
s
y
s
tem
W
h
ile
th
e
q
u
ality
o
f
k
n
o
wle
d
g
e
r
ep
r
esen
tatio
n
[
1
4
]
,
t
h
e
ef
f
icac
y
o
f
r
etr
iev
al
[
1
5
]
–
[
1
7
]
,
an
d
t
h
e
p
er
f
o
r
m
an
ce
o
f
L
L
Ms
[
1
8
]
h
av
e
b
ee
n
th
e
f
o
c
u
s
o
f
ex
ten
s
iv
e
r
esear
ch
in
th
e
R
AG
f
ield
,
th
e
p
r
e
-
in
d
e
x
ch
u
n
k
i
n
g
s
tep
r
em
ain
s
a
p
iv
o
tal
f
ac
to
r
.
T
h
is
s
tep
d
ef
in
es
t
h
e
g
r
an
u
lar
ity
o
f
in
d
ex
e
d
u
n
it
s
an
d
co
n
s
eq
u
en
tly
in
f
lu
en
ce
s
b
o
th
r
etr
iev
al
ac
cu
r
ac
y
an
d
g
e
n
er
atio
n
q
u
ality
.
T
h
e
s
elec
tio
n
o
f
ch
u
n
k
in
g
is
a
cr
itical
f
ac
to
r
in
th
e
ef
f
icac
y
o
f
i
n
f
o
r
m
atio
n
r
etr
i
ev
al.
Po
o
r
ly
c
h
o
s
en
c
h
u
n
k
in
g
ca
n
r
esu
lt
in
th
e
f
r
a
g
m
en
tatio
n
o
f
co
h
er
en
t
in
f
o
r
m
atio
n
o
r
,
co
n
v
er
s
ely
,
t
h
e
o
v
er
lo
a
d
in
g
o
f
th
e
r
etr
iev
er
with
o
v
er
ly
b
r
o
ad
s
eg
m
e
n
ts
.
I
n
b
o
th
ca
s
es,
d
o
wn
s
tr
ea
m
p
er
f
o
r
m
a
n
ce
is
n
eg
ativ
ely
im
p
ac
ted
.
D
o
c
u
m
e
n
t
c
h
u
n
k
i
n
g
s
t
r
a
t
e
g
i
es
h
a
v
e
e
v
o
l
v
e
d
f
r
o
m
s
i
m
p
l
e
h
e
u
r
i
s
t
i
cs
t
o
s
o
p
h
i
s
t
i
c
at
e
d
L
L
M
-
d
r
i
v
e
n
a
p
p
r
o
a
c
h
e
s
.
T
a
b
l
e
1
p
r
e
s
e
n
ts
a
t
a
x
o
n
o
m
y
o
f
t
h
e
m
a
i
n
c
h
u
n
k
i
n
g
s
t
r
a
t
e
g
ie
s
,
s
u
m
m
a
r
i
z
i
n
g
th
e
i
r
c
o
r
e
p
r
i
n
c
i
p
l
e
s
,
s
t
r
e
n
g
t
h
s
,
a
n
d
l
i
m
it
a
t
i
o
n
s
.
A
d
et
a
i
l
e
d
e
x
p
l
a
n
at
i
o
n
o
f
e
a
c
h
c
h
u
n
k
i
n
g
s
t
r
a
t
e
g
y
is
p
r
o
v
i
d
e
d
i
n
t
h
e
n
e
x
t
p
a
r
a
g
r
a
p
h
s
.
T
ab
le
1
.
T
a
x
o
n
o
m
y
o
f
ch
u
n
k
i
n
g
s
tr
ateg
ies f
o
r
R
AG
Fix
ed
-
s
ize
ch
ar
ac
ter
ch
u
n
k
i
n
g
r
e
p
r
esen
ts
th
e
f
o
u
n
d
atio
n
al
m
eth
o
d
,
d
iv
id
in
g
d
o
cu
m
en
ts
in
to
co
n
tig
u
o
u
s
win
d
o
ws
o
f
p
r
ed
e
ter
m
in
ed
c
h
ar
ac
ter
len
g
th
[
2
]
.
R
ec
u
r
s
iv
e
ch
u
n
k
in
g
b
u
ild
s
u
p
o
n
th
is
f
o
u
n
d
atio
n
C
h
u
n
k
i
n
g
s
t
r
a
t
e
g
y
D
e
scri
p
t
i
o
n
S
t
r
e
n
g
t
h
s
W
e
a
k
n
e
sses
F
i
x
e
d
-
si
z
e
c
h
a
r
a
c
t
e
r
c
h
u
n
k
i
n
g
S
p
l
i
t
s
t
e
x
t
i
n
t
o
u
n
i
f
o
r
m se
g
m
e
n
t
s
b
y
c
h
a
r
a
c
t
e
r
c
o
u
n
t
(
e
.
g
.
,
e
v
e
r
y
1
,
0
0
0
c
h
a
r
a
c
t
e
r
s)
S
i
mp
l
e
a
n
d
f
a
st
t
o
i
mp
l
e
m
e
n
t
B
r
e
a
k
s
c
o
n
t
e
x
t
a
n
d
i
g
n
o
r
e
s c
o
n
t
e
n
t
b
o
u
n
d
a
r
i
e
s
R
e
c
u
r
si
v
e
c
h
u
n
k
i
n
g
S
p
l
i
t
s
b
y
s
t
r
u
c
t
u
r
a
l
mar
k
e
r
s (se
c
t
i
o
n
s
,
p
a
r
a
g
r
a
p
h
s,
a
n
d
l
i
n
e
s)
,
u
si
n
g
a
r
e
c
u
r
si
v
e
h
i
e
r
a
r
c
h
y
o
f
s
e
p
a
r
a
t
o
r
s
P
r
e
serv
e
s
d
o
c
u
m
e
n
t
st
r
u
c
t
u
r
e
,
b
e
t
t
e
r
c
o
n
t
e
x
t
S
o
me
t
i
m
e
s
p
r
o
d
u
c
e
s
u
n
e
v
e
n
si
z
e
s
S
e
ma
n
t
i
c
c
h
u
n
k
i
n
g
C
h
u
n
k
s
a
r
e
d
e
t
e
r
m
i
n
e
d
u
si
n
g
sem
a
n
t
i
c
si
mi
l
a
r
i
t
y
,
e
mb
e
d
d
i
n
g
s,
o
r
t
o
p
i
c
c
h
a
n
g
e
s
;
mai
n
t
a
i
n
s m
e
a
n
i
n
g
c
o
n
t
i
n
u
i
t
y
P
r
o
d
u
c
e
s c
o
h
e
r
e
n
t
,
mea
n
i
n
g
f
u
l
c
h
u
n
k
s
t
h
a
t
i
mp
r
o
v
e
r
e
t
r
i
e
v
a
l
C
o
m
p
u
t
a
t
i
o
n
a
l
l
y
i
n
t
e
n
si
v
e
LLM
-
b
a
s
e
d
c
h
u
n
k
i
n
g
U
t
i
l
i
z
e
s a
n
LL
M
t
o
i
n
t
e
l
l
i
g
e
n
t
l
y
se
g
m
e
n
t
d
o
c
u
me
n
t
s
a
t
sem
a
n
t
i
c
a
l
l
y
a
p
p
r
o
p
r
i
a
t
e
b
o
u
n
d
a
r
i
e
s
C
r
e
a
t
e
s
h
i
g
h
l
y
s
e
ma
n
t
i
c
a
n
d
c
o
n
t
e
x
t
-
r
i
c
h
c
h
u
n
k
s
C
o
m
p
u
t
a
t
i
o
n
a
l
l
y
e
x
p
e
n
si
v
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
9
0
9
-
1
9
1
8
1912
b
y
in
c
o
r
p
o
r
atin
g
s
tr
u
ct
u
r
al
awa
r
en
ess
—
ap
p
ly
in
g
s
ize
c
o
n
s
tr
ain
ts
wh
ile
p
r
eser
v
in
g
n
atu
r
al
d
o
cu
m
e
n
t
b
o
u
n
d
ar
ies
s
u
ch
as
h
ea
d
in
g
s
,
p
ar
ag
r
ap
h
s
,
an
d
s
en
ten
ce
b
r
ea
k
s
.
Mo
r
e
s
o
p
h
is
ticated
ap
p
r
o
ac
h
es
f
o
cu
s
o
n
s
em
an
tic
co
h
er
e
n
ce
.
Sem
a
n
tic
ch
u
n
k
in
g
id
en
tifie
s
o
p
tim
al
s
eg
m
en
tatio
n
p
o
i
n
ts
b
y
d
ete
ctin
g
to
p
ical
s
h
if
ts
,
ty
p
ically
th
r
o
u
g
h
c
o
n
tin
u
o
u
s
m
o
n
ito
r
in
g
o
f
s
em
an
tic
s
im
ilar
ity
b
etwe
en
ad
jace
n
t
s
en
t
en
ce
s
,
an
d
p
lacin
g
b
o
u
n
d
ar
ies
wh
er
e
co
h
er
e
n
ce
f
alls
b
elo
w
a
d
ef
in
ed
th
r
esh
o
ld
[
1
9
]
–
[
2
1
]
.
Nev
er
t
h
eless
,
r
ec
en
t
s
tu
d
ies
[
2
1
]
q
u
esti
o
n
s
em
an
tic
c
h
u
n
k
in
g
'
s
ef
f
ec
tiv
en
ess
g
iv
en
its
s
u
b
s
tan
tial
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
,
wh
ich
r
eq
u
ir
es
b
o
th
em
b
ed
d
in
g
g
e
n
er
atio
n
a
n
d
s
i
m
ilar
ity
ca
lcu
latio
n
s
f
o
r
e
v
er
y
p
o
ten
tial b
o
u
n
d
a
r
y
d
ec
is
io
n
.
Giv
en
L
L
Ms’
d
em
o
n
s
tr
ated
ex
ce
llen
ce
ac
r
o
s
s
d
iv
er
s
e
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
task
s
,
r
ec
en
t
r
esear
ch
h
as
in
cr
ea
s
in
g
ly
f
o
c
u
s
ed
o
n
lev
er
ag
in
g
th
ese
m
o
d
els
d
ir
ec
tly
f
o
r
in
tellig
en
t
d
o
cu
m
en
t
ch
u
n
k
in
g
.
LLM
-
b
ased
ch
u
n
k
in
g
ap
p
r
o
ac
h
es
h
ar
n
ess
th
e
r
ea
s
o
n
in
g
ca
p
ab
ilit
ies
o
f
L
L
Ms
to
m
ak
e
in
tellig
en
t
s
eg
m
en
tatio
n
d
ec
is
io
n
s
.
T
h
ese
r
an
g
e
f
r
o
m
s
tr
aig
h
tf
o
r
war
d
p
r
o
m
p
tin
g
s
tr
ateg
ies to
ad
v
an
ce
d
ar
ch
itectu
r
es lik
e
th
e
m
ix
tu
r
e
-
of
-
ch
u
n
k
er
s
(
Mo
C
)
m
eth
o
d
[
2
2
]
,
wh
ich
em
p
lo
y
s
a
r
o
u
tin
g
m
ec
h
a
n
is
m
to
d
y
n
am
ically
s
elec
t
th
e
m
o
s
t
ap
p
r
o
p
r
iate
g
r
an
u
lar
ity
f
o
r
d
if
f
er
e
n
t
d
o
cu
m
e
n
t
s
ec
tio
n
s
,
o
r
L
u
m
b
e
r
C
h
u
n
k
e
r
[
2
3
]
,
wh
ich
em
p
lo
y
s
L
L
Ms
to
d
y
n
a
m
ically
s
eg
m
en
t d
o
cu
m
en
ts
in
to
s
em
an
tically
in
d
ep
en
d
en
t c
h
u
n
k
s
.
R
ec
en
t
r
esear
ch
h
as
i
n
cr
ea
s
in
g
ly
f
o
cu
s
ed
o
n
s
y
s
tem
atic
e
v
alu
atio
n
o
f
c
h
u
n
k
in
g
s
tr
ateg
ies
ac
r
o
s
s
d
iv
er
s
e
co
n
tex
ts
an
d
ap
p
licatio
n
s
.
T
h
e
Om
n
iEv
al
b
en
c
h
m
ar
k
[
7
]
estab
lis
h
ed
a
c
o
m
p
r
e
h
en
s
iv
e
ev
alu
atio
n
f
r
am
ewo
r
k
f
o
r
th
e
f
in
an
cial
d
o
m
ain
,
s
y
s
tem
atica
lly
v
ar
y
i
n
g
ch
u
n
k
s
izes,
o
v
er
lap
p
ar
a
m
eter
s
,
em
b
ed
d
in
g
m
o
d
els,
an
d
r
etr
iev
al
m
ec
h
a
n
is
m
s
to
ass
ess
th
eir
co
m
b
in
ed
im
p
ac
t
o
n
p
er
f
o
r
m
a
n
ce
.
B
u
ild
in
g
o
n
th
is
f
o
u
n
d
atio
n
,
th
e
M
o
C
ap
p
r
o
ac
h
[
2
2
]
d
em
o
n
s
tr
ated
th
at
in
te
llig
en
tly
r
o
u
tin
g
b
etwe
en
m
u
l
tip
le
s
eg
m
en
tatio
n
g
r
an
u
lar
ities
ca
n
s
ig
n
if
ica
n
tly
en
h
an
ce
d
o
wn
s
tr
ea
m
R
AG
p
e
r
f
o
r
m
a
n
ce
.
T
h
e
f
ield
h
as
s
in
ce
e
x
p
an
d
ed
b
ey
o
n
d
p
u
r
e
s
eg
m
en
tatio
n
o
p
tim
izatio
n
.
C
h
u
n
k
R
AG
[
2
4
]
s
h
if
ts
th
e
p
ar
ad
ig
m
b
y
in
tr
o
d
u
cin
g
L
L
M
-
b
ased
p
o
s
t
-
r
etr
iev
al
f
ilter
i
n
g
,
f
o
c
u
s
in
g
o
n
r
ef
in
in
g
an
d
cu
r
atin
g
r
etr
iev
e
d
ch
u
n
k
s
r
at
h
er
th
an
p
er
f
ec
tin
g
in
itial
s
eg
m
en
tatio
n
—
s
p
ec
if
ically
tar
g
etin
g
th
e
r
em
o
v
al
o
f
ir
r
elev
a
n
t
o
r
r
ed
u
n
d
an
t
co
n
tex
t
b
ef
o
r
e
an
s
wer
g
en
er
atio
n
.
Me
an
wh
ile
,
co
m
p
ar
ativ
e
s
tu
d
ies
[
9
]
h
av
e
ex
am
in
ed
th
e
tr
ad
e
-
o
f
f
s
b
etwe
en
late
ch
u
n
k
in
g
[
2
5
]
an
d
co
n
tex
t
u
al
r
etr
iev
al
[
2
6
]
,
r
e
v
ea
lin
g
th
at
wh
ile
co
n
tex
tu
al
r
etr
iev
al
b
etter
p
r
eser
v
es
s
em
an
tic
co
h
er
en
ce
,
it
c
o
m
es
with
s
u
b
s
tan
t
ial
co
m
p
u
tatio
n
al
o
v
e
r
h
ea
d
.
L
ate
ch
u
n
k
in
g
o
f
f
e
r
s
g
r
ea
ter
ef
f
icien
c
y
b
u
t o
f
ten
at
th
e
ex
p
en
s
e
o
f
r
etr
iev
al
r
elev
a
n
ce
an
d
a
n
s
wer
co
m
p
leten
ess
.
3.
E
XP
E
R
I
M
E
N
T
S
T
h
is
s
ec
tio
n
p
r
o
v
i
d
es
th
e
ex
p
er
im
en
tal
d
etails
o
f
o
u
r
s
tu
d
y
,
in
clu
d
in
g
th
e
d
ataset
co
n
s
tr
u
ctio
n
an
d
th
e
ev
alu
atio
n
p
r
o
t
o
co
l.
Su
b
-
s
ec
tio
n
3
.
1
d
escr
ib
es
th
e
cr
ea
ti
o
n
o
f
a
h
ig
h
-
q
u
ality
,
d
o
m
ain
-
s
p
ec
if
ic
co
r
p
u
s
an
d
th
e
g
en
er
atio
n
o
f
g
r
o
u
n
d
-
t
r
u
th
QA
p
air
s
u
s
ed
f
o
r
ev
alu
atio
n
.
Su
b
-
s
ec
tio
n
3
.
2
d
etails
th
e
ev
alu
atio
n
f
r
am
ewo
r
k
,
d
etailin
g
th
e
r
etr
i
ev
al
an
d
g
e
n
er
atio
n
m
etr
ics,
alo
n
g
with
th
e
c
o
n
f
i
g
u
r
atio
n
s
u
s
ed
to
ass
ess
th
e
p
er
f
o
r
m
an
ce
o
f
v
ar
i
o
u
s
ch
u
n
k
in
g
s
tr
ateg
ies with
in
th
e
R
AG
p
ip
elin
e.
3
.
1
.
Da
t
a
s
et
W
e
ass
em
b
led
a
d
o
m
ain
-
s
p
ec
i
f
ic
co
r
p
u
s
b
y
cu
r
atin
g
h
ig
h
-
q
u
ality
ed
ito
r
ial
ar
ticles
o
n
SaaS
s
o
f
twar
e
f
r
o
m
two
estab
lis
h
ed
m
ed
ia
s
o
u
r
ce
s
:
Ap
p
v
izer
(
h
ttp
s
://www.
ap
p
v
izer
.
co
m
/)
a
n
d
B
lo
g
d
u
Mo
d
ér
ateu
r
(
h
ttp
s
://www.
b
lo
g
d
u
m
o
d
er
ate
u
r
.
co
m
/)
.
Af
ter
co
llectin
g
an
d
n
o
r
m
alizin
g
th
e
r
aw
tex
t
f
r
o
m
th
e
ca
n
d
id
ate
web
p
ag
es,
we
co
n
s
tr
u
cted
a
g
r
o
u
n
d
-
tr
u
t
h
QA
d
ataset
f
o
llo
win
g
a
s
tr
u
ctu
r
ed
g
e
n
er
a
tio
n
an
d
v
alid
atio
n
p
r
o
ce
s
s
[
8
]
,
[
2
7
]
.
E
ac
h
ar
ticle
was
f
ir
s
t
s
eg
m
en
ted
in
to
r
elativ
ely
lar
g
e
p
ass
ag
es,
wh
ich
wer
e
th
en
u
s
ed
as
co
n
tex
t
in
p
r
o
m
p
ts
f
o
r
an
L
L
M
(
GPT
-
4
o
in
o
u
r
ca
s
e)
to
g
e
n
er
ate
co
n
cise,
f
ac
t
-
s
ee
k
in
g
q
u
esti
o
n
s
p
air
ed
with
th
eir
co
r
r
esp
o
n
d
i
n
g
a
n
s
wer
s
,
b
o
th
s
tr
ictly
g
r
o
u
n
d
ed
in
th
e
s
o
u
r
ce
p
ass
ag
e.
T
o
en
s
u
r
e
q
u
ality
,
th
e
r
esu
ltin
g
QA
p
air
s
wer
e
e
v
alu
ated
b
y
GPT
-
4
o
its
elf
as
a
ju
d
g
e,
wh
ich
r
ated
th
eir
clar
ity
,
f
ac
tu
al
ac
cu
r
ac
y
,
a
n
d
an
s
wer
ab
ilit
y
o
n
a
f
iv
e
-
p
o
in
t
s
ca
le.
On
ly
p
air
s
r
ec
eiv
in
g
s
c
o
r
es
ab
o
v
e
f
o
u
r
o
u
t
o
f
f
iv
e
o
n
all
cr
iter
ia
wer
e
r
etain
ed
.
As
a
r
esu
lt,
2
,
4
1
9
QA
p
air
s
,
ea
c
h
alig
n
ed
wit
h
its
co
r
r
esp
o
n
d
i
n
g
p
ass
ag
e,
wer
e
r
etain
e
d
f
o
r
ev
alu
atio
n
.
T
h
is
p
ip
elin
e
y
iel
d
ed
a
s
ca
lab
le,
h
ig
h
-
q
u
ality
QA
r
eso
u
r
ce
tailo
r
e
d
to
th
e
p
u
b
lis
h
in
g
i
n
d
u
s
tr
y
r
eg
ar
d
in
g
SaaS,
p
r
o
v
id
in
g
a
r
eliab
le
b
asis
f
o
r
th
e
s
y
s
tem
atic
ev
alu
atio
n
o
f
ch
u
n
k
in
g
s
tr
ateg
ies
in
R
AG.
Fig
u
r
e
2
p
r
esen
ts
an
ex
am
p
l
e
o
f
a
p
ass
ag
e
–
q
u
esti
o
n
–
a
n
s
wer
tr
ip
le.
E
ac
h
p
ass
ag
e
was
u
s
ed
to
g
en
er
ate
a
f
ac
t
-
s
ee
k
in
g
q
u
esti
o
n
an
d
its
co
r
r
esp
o
n
d
in
g
a
n
s
wer
,
b
o
t
h
g
r
o
u
n
d
ed
i
n
th
e
s
o
u
r
ce
p
ass
ag
e
3
.
2
.
E
v
a
lua
t
i
o
n pro
t
o
co
l
W
e
ev
alu
ated
t
h
e
ca
n
d
id
ate
c
h
u
n
k
in
g
s
tr
ateg
ies
ac
r
o
s
s
b
o
t
h
th
e
r
etr
iev
al
an
d
g
en
er
atio
n
p
h
ases
o
f
th
e
R
AG
p
ip
elin
e.
Ou
r
p
r
o
to
co
l
co
m
b
in
es
tr
a
d
itio
n
al
s
im
il
ar
ity
-
b
ased
m
etr
ics
with
L
L
M
-
b
ased
ju
d
g
m
en
ts
,
th
er
eb
y
ca
p
tu
r
in
g
co
m
p
lem
en
t
ar
y
asp
ec
ts
o
f
p
er
f
o
r
m
an
ce
.
T
h
is
co
m
p
r
eh
en
s
iv
e
ev
alu
atio
n
d
esig
n
en
s
u
r
es
th
at
b
o
th
t
h
e
q
u
an
titativ
e
p
r
ec
is
io
n
o
f
r
etr
ie
v
al
an
d
th
e
q
u
alitativ
e
r
ea
s
o
n
in
g
a
b
ilit
y
o
f
th
e
g
en
er
ativ
e
m
o
d
el
a
r
e
s
y
s
tem
atica
lly
ass
e
s
s
ed
with
in
a
u
n
if
ied
e
x
p
er
im
e
n
tal
f
r
am
e
wo
r
k
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
va
lu
a
tin
g
d
o
cu
me
n
t c
h
u
n
kin
g
a
p
p
r
o
a
ch
es fo
r
r
etri
ev
a
l
-
a
u
g
men
ted
g
e
n
era
tio
n
in
…
(
E
r
w
a
n
n
La
va
r
ec
)
1913
Fig
u
r
e
2
.
I
ll
u
s
tr
atio
n
o
f
a
p
ass
ag
e
–
q
u
esti
o
n
–
an
s
wer
tr
ip
le
3
.
2
.
1
.
Ret
riev
a
l e
v
a
lua
t
io
n
W
e
ass
e
s
s
h
o
w
ef
f
ec
tiv
ely
r
et
r
iev
ed
d
o
cu
m
e
n
t
ch
u
n
k
s
alig
n
with
th
e
g
o
ld
r
ef
er
e
n
ce
p
ass
ag
e
u
s
ed
t
o
co
n
s
tr
u
ct
th
e
co
r
r
esp
o
n
d
in
g
QA
p
air
.
T
h
e
g
en
er
al
p
u
r
p
o
s
e
o
f
th
is
ev
alu
atio
n
is
to
m
ea
s
u
r
e
th
e
ca
p
ab
ilit
y
o
f
id
en
tify
in
g
r
elev
an
t c
o
n
tex
t to
th
e
q
u
esti
o
n
f
o
r
th
e
L
L
M.
T
h
r
ee
m
etr
ics we
r
e
u
s
ed
:
i)
Simp
le
co
n
tex
t
p
r
ec
is
io
n
:
th
is
m
etr
ic
m
ea
s
u
r
es
th
e
p
r
o
p
o
r
ti
o
n
o
f
r
etr
iev
e
d
ch
u
n
k
s
th
at
o
r
ig
in
ate
f
r
o
m
th
e
s
am
e
s
o
u
r
ce
r
aw
d
o
cu
m
e
n
t
as
th
e
g
o
ld
r
e
f
er
en
ce
p
ass
ag
e.
I
t
d
ir
ec
tly
r
ef
lects
th
e
d
o
cu
m
en
t
-
lev
el
lo
ca
lizatio
n
ca
p
ab
ilit
y
o
f
th
e
r
etr
iev
er
.
ii)
Sem
an
tic
s
im
ilar
ity
:
th
is
m
etr
ic
co
m
p
u
tes
th
e
co
s
in
e
s
im
ilar
ity
b
etwe
en
its
em
b
ed
d
in
g
a
n
d
th
e
em
b
ed
d
in
g
o
f
th
e
g
o
ld
p
ass
ag
e.
T
h
is
m
etr
ic
ca
p
tu
r
es
s
em
an
tic
clo
s
en
ess
b
etwe
en
th
e
r
etr
i
ev
ed
c
o
n
tex
t
an
d
th
e
s
o
u
r
ce
p
ass
ag
e.
Op
en
AI
’
s
tex
t
-
em
b
ed
d
in
g
-
3
-
lar
g
e
m
o
d
el
was u
s
ed
.
iii)
LLM
-
b
ased
co
n
tex
t
p
r
ec
is
io
n
:
th
is
m
etr
ic
u
s
es
an
L
L
M
ev
alu
ato
r
to
ju
d
g
e
wh
eth
er
e
ac
h
r
etr
iev
ed
co
n
tex
t
is
r
elev
an
t
to
an
s
wer
i
n
g
th
e
g
iv
e
n
q
u
esti
o
n
,
r
eg
ar
d
less
o
f
d
o
cu
m
en
t
p
r
o
v
e
n
an
ce
.
T
h
is
m
etr
ic
ac
co
u
n
ts
f
o
r
p
r
a
g
m
atic
u
s
ef
u
ln
ess
an
d
ca
n
h
ig
h
lig
h
t
c
ases
wer
e
tr
ad
itio
n
al
s
im
ilar
ity
m
ea
s
u
r
es
u
n
d
er
esti
m
ate
r
etr
iev
al
q
u
ality
.
I
n
o
u
r
ex
p
e
r
im
en
ts
,
GPT
-
4
o
was
u
s
ed
as
th
e
ev
alu
ato
r
.
T
h
e
im
p
lem
en
tatio
n
o
f
th
is
m
etr
i
c
f
o
llo
wed
th
e
r
etr
iev
al
au
g
m
en
ted
g
en
er
atio
n
ass
ess
m
en
t
s
(
R
AGAs
)
f
r
am
ewo
r
k
[
2
8
]
,
wh
ich
p
r
o
v
id
es st
an
d
ar
d
ized
to
o
ls
f
o
r
L
L
M
-
b
ased
R
AG
ass
es
s
m
en
t.
3
.
2
.
2
.
G
ener
a
t
io
n e
v
a
lua
t
io
n
W
e
f
u
r
th
er
ev
alu
ate
th
e
e
n
d
-
to
-
en
d
ca
p
ac
ity
o
f
t
h
e
R
AG
s
y
s
tem
to
p
r
o
d
u
ce
co
r
r
ec
t
a
n
d
i
n
f
o
r
m
ativ
e
an
s
wer
s
.
Fo
r
th
is
,
we
em
p
lo
y
GPT
-
4
o
as
a
ju
d
g
e.
Usi
n
g
a
C
o
T
p
r
o
m
p
tin
g
s
tr
ateg
y
[
2
9
]
,
th
e
m
o
d
el
is
f
ir
s
t
in
s
tr
u
cted
to
p
r
o
v
id
e
s
tr
u
ctu
r
ed
f
ee
d
b
ac
k
b
y
co
m
p
ar
in
g
th
e
g
en
er
ate
d
a
n
s
wer
with
b
o
th
th
e
in
p
u
t
q
u
esti
o
n
an
d
th
e
g
o
ld
r
ef
e
r
en
ce
an
s
we
r
.
I
t
th
en
ass
ig
n
s
a
q
u
ality
s
c
o
r
e
o
n
a
f
iv
e
-
p
o
in
t
s
ca
le,
r
ef
l
ec
tin
g
th
e
an
s
wer
’
s
ac
cu
r
ac
y
,
f
ac
tu
al
c
o
r
r
ec
tn
ess
,
an
d
c
o
m
p
leten
ess
.
T
h
e
s
co
r
i
n
g
r
u
b
r
ic,
d
ef
in
i
n
g
a
n
ch
o
r
cr
i
ter
ia
f
o
r
ea
ch
s
co
r
e
f
r
o
m
1
to
5
,
is
p
r
o
v
id
ed
as
f
o
l
lo
ws
to
g
eth
er
with
th
e
f
u
ll
ev
alu
atio
n
p
r
o
m
p
t
tem
p
late
u
s
ed
in
o
u
r
ex
p
e
r
im
en
ts
to
en
s
u
r
e
tr
an
s
p
a
r
en
cy
a
n
d
r
ep
r
o
d
u
cib
ilit
y
.
Pro
m
p
t te
m
p
late
u
s
ed
i
n
L
L
M
-
as
-
a
-
ju
d
g
e
f
o
r
g
en
er
atio
n
ev
a
lu
atio
n
#
#
#
T
ask
Descr
ip
tio
n
:
An
in
s
tr
u
ctio
n
,
a
r
esp
o
n
s
e
to
ev
alu
ate,
a
r
ef
er
e
n
ce
an
s
wer
,
an
d
a
s
co
r
e
r
u
b
r
ic
r
ep
r
esen
tin
g
ev
alu
atio
n
cr
iter
ia
ar
e
g
iv
en
.
1
.
W
r
ite
d
etailed
f
ee
d
b
ac
k
t
h
a
t
ass
es
s
th
e
q
u
ality
o
f
th
e
r
esp
o
n
s
e
s
tr
ictly
b
ased
o
n
th
e
g
iv
e
n
s
co
r
e
r
u
b
r
ic,
n
o
t
ev
alu
atin
g
in
g
en
er
al.
2
.
Af
ter
wr
itin
g
a
f
ee
d
b
ac
k
,
a
s
s
ig
n
a
s
co
r
e
th
at
is
an
in
teg
er
b
etwe
en
1
an
d
5
.
Y
o
u
s
h
o
u
l
d
r
ef
er
to
t
h
e
s
co
r
e
r
u
b
r
ic.
3
.
T
h
e
o
u
tp
u
t
f
o
r
m
at
s
h
o
u
ld
b
e
as
f
o
llo
ws:
"Fee
d
b
ac
k
:
{{
wr
ite
y
o
u
r
f
ee
d
b
ac
k
f
o
r
cr
iter
ia}
}
[
R
E
SULT
]
{{
an
in
teg
er
n
u
m
b
er
b
etwe
en
1
a
n
d
5
}}
"
4
.
Please
d
o
n
o
t
g
en
er
ate
an
y
o
th
er
o
p
en
i
n
g
,
clo
s
in
g
,
an
d
ex
p
lan
atio
n
s
.
B
e
s
u
r
e
to
in
clu
d
e
[
R
E
SULT
]
in
y
o
u
r
o
u
tp
u
t.
#
#
#
T
h
e
in
s
tr
u
ctio
n
:
{in
s
tr
u
ctio
n
}
#
#
#
R
esp
o
n
s
e
to
ev
al
u
ate:
{r
esp
o
n
s
e}
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
9
0
9
-
1
9
1
8
1914
#
#
#
R
ef
er
en
ce
A
n
s
wer
:
{r
ef
er
en
ce
_
a
n
s
wer
}
#
#
#
Sco
r
e
R
u
b
r
ics:
[
I
s
th
e
r
esp
o
n
s
e
co
r
r
ec
t,
ac
c
u
r
ate,
an
d
f
ac
tu
al
b
ased
o
n
th
e
r
ef
er
en
ce
an
s
wer
?]
Sco
r
e
1
: T
h
e
r
esp
o
n
s
e
is
co
m
p
letely
in
co
r
r
ec
t,
in
ac
c
u
r
ate,
a
n
d
/o
r
n
o
t f
ac
tu
al.
Sco
r
e
2
: T
h
e
r
esp
o
n
s
e
is
m
o
s
tly
in
co
r
r
ec
t,
in
ac
cu
r
ate
,
an
d
/
o
r
n
o
t f
ac
tu
al.
Sco
r
e
3
: T
h
e
r
esp
o
n
s
e
is
s
o
m
e
wh
at
co
r
r
ec
t,
ac
c
u
r
ate,
an
d
/o
r
f
ac
tu
al.
Sco
r
e
4
: T
h
e
r
esp
o
n
s
e
is
m
o
s
tly
co
r
r
ec
t,
ac
cu
r
ate,
a
n
d
f
ac
t
u
al.
Sco
r
e
5
: T
h
e
r
esp
o
n
s
e
is
co
m
p
letely
co
r
r
ec
t,
ac
cu
r
ate,
an
d
f
a
ctu
al.
#
#
#
Feed
b
ac
k
:
3
.
2
.
3
.
Chun
k
er
co
nfig
ura
t
io
ns
F
o
r
th
e
ev
a
l
u
a
t
i
o
n
,
w
e
a
n
al
y
z
e
d
f
o
u
r
d
i
s
t
in
c
t
d
o
cu
m
en
t
c
h
u
n
k
i
n
g
s
t
r
a
t
e
g
i
e
s
.
T
h
e
b
a
s
e
l
i
n
e
a
p
p
r
o
a
ch
e
s
c
o
n
s
i
s
t
e
d
o
f
f
i
x
e
d
-
s
i
z
e
ch
a
r
a
c
te
r
c
h
u
n
k
i
n
g
a
n
d
r
ec
.
u
r
s
i
v
e
ch
u
n
k
in
g
,
b
o
t
h
o
f
w
h
i
ch
w
e
r
e
e
v
a
l
u
a
t
e
d
a
t
i
n
te
r
v
a
l
s
o
f
5
0
0
,
7
5
0
,
1
,
0
0
0
,
1
,
2
5
0
,
a
n
d
1
,
5
0
0
ch
ar
a
c
t
er
s
.
F
o
r
m
o
r
e
c
o
n
t
e
x
t
-
a
w
a
r
e
s
e
g
m
e
n
ta
t
i
o
n
,
w
e
u
t
i
l
i
z
ed
s
em
a
n
t
i
c
c
h
u
n
k
i
n
g
w
i
t
h
a
p
er
c
en
t
i
l
e
-
b
a
s
e
d
b
r
ea
k
p
o
in
t
t
h
r
e
s
h
o
l
d
o
f
0
.
9
5
.
F
i
n
a
l
ly
,
w
e
i
n
co
r
p
o
r
a
t
ed
L
L
M
-
b
a
s
e
d
c
h
u
n
k
i
n
g
im
p
l
em
e
n
te
d
v
i
a
G
P
T
-
4
o
,
te
s
t
i
n
g
b
o
t
h
a
s
t
a
n
d
a
r
d
b
a
s
i
c
ap
p
r
o
a
ch
[
2
2
]
an
d
a
Mo
C
v
ar
i
a
n
t
[
2
2
]
.
4.
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
em
p
ir
ical
r
esu
lts
o
f
o
u
r
s
tu
d
y
,
c
o
m
p
ar
in
g
c
h
u
n
k
in
g
s
tr
ateg
ies
ac
r
o
s
s
b
o
th
r
etr
iev
al
an
d
g
en
er
atio
n
m
etr
i
cs.
T
h
e
an
aly
s
is
h
ig
h
lig
h
ts
h
o
w
d
if
f
er
en
t
s
eg
m
en
tatio
n
ap
p
r
o
ac
h
es
in
f
lu
en
ce
r
etr
iev
al
p
r
ec
is
io
n
,
s
em
an
tic
s
im
ilar
ity
,
an
d
en
d
-
to
-
en
d
an
s
wer
q
u
ality
.
I
n
a
d
d
itio
n
,
th
e
d
is
cu
s
s
io
n
in
ter
p
r
ets
th
ese
f
in
d
in
g
s
to
id
e
n
tify
tr
a
d
e
-
o
f
f
s
b
etwe
en
c
h
u
n
k
s
ize,
s
tr
u
ctu
r
al
awa
r
en
ess
,
an
d
o
v
er
all
R
AG
p
er
f
o
r
m
an
ce
.
Fig
u
r
e
3
i
ll
u
s
t
r
at
es th
e
im
p
ac
t
o
f
ch
u
n
k
s
ize
o
n
r
e
tr
ie
v
al
p
e
r
f
o
r
m
a
n
ce
f
o
r
t
h
e
c
h
a
r
a
ct
er
-
b
ase
d
m
e
t
h
o
d
s
(
f
ix
e
d
-
s
i
ze
a
n
d
r
e
c
u
r
s
i
v
e)
.
As
d
e
m
o
n
s
tr
ate
d
i
n
t
h
e
f
o
l
lo
wi
n
g
i
n
v
esti
g
ati
o
n
,
f
o
r
th
e
s
i
m
p
l
e
c
o
n
te
x
t
p
r
ec
is
i
o
n
m
et
r
i
c,
s
m
all
er
c
h
u
n
k
s
c
o
n
s
is
ten
tl
y
y
i
el
d
h
i
g
h
er
d
o
c
u
m
en
t
-
lev
el
l
o
ca
liz
ati
o
n
.
M
o
r
e
o
v
er
,
r
e
cu
r
s
i
v
e
c
h
u
n
k
i
n
g
o
u
t
p
e
r
f
o
r
m
s
f
i
x
e
d
-
s
iz
e
wi
n
d
o
ws
at
a
ll
s
c
ales
.
Se
m
a
n
t
ic
s
i
m
i
lar
it
y
h
as
b
ee
n
s
h
o
w
n
to
in
c
r
e
ase
wi
th
ch
u
n
k
s
iz
e,
r
e
ac
h
i
n
g
a
p
ea
k
a
t
a
p
p
r
o
x
i
m
a
tel
y
1
,
2
5
0
-
1
,
5
0
0
c
h
a
r
ac
te
r
s
.
T
h
is
s
u
g
g
ests
t
h
at
l
ar
g
e
r
s
e
g
m
e
n
ts
m
a
y
p
r
es
e
r
v
e
m
o
r
e
t
o
p
i
ca
l
co
n
t
ex
t.
Fi
n
all
y
,
L
L
M
-
b
as
ed
co
n
t
ex
t p
r
e
cisi
o
n
u
n
d
er
s
co
r
es a
tr
a
d
e
-
o
f
f
:
r
ec
u
r
s
iv
e
ch
u
n
k
i
n
g
att
ai
n
s
o
p
ti
m
al
p
e
r
f
o
r
m
a
n
c
e
at
in
te
r
m
ed
iat
e
s
iz
es
(
a
p
p
r
o
x
im
at
el
y
7
5
0
c
h
a
r
ac
t
er
s
)
,
w
h
ile
f
i
x
e
d
-
s
iz
e
c
h
u
n
k
i
n
g
m
ai
n
t
ai
n
s
g
r
e
ate
r
s
ta
b
i
lit
y
a
cr
o
s
s
th
e
s
p
ec
t
r
u
m
.
T
h
es
e
c
o
m
p
lem
en
ta
r
y
p
a
tte
r
n
s
i
n
d
ica
te
t
h
a
t
t
h
e
ch
u
n
k
s
ize
h
as
a
d
i
r
e
ct
im
p
a
ct
o
n
t
h
e
e
q
u
ili
b
r
i
u
m
b
et
wee
n
f
in
e
-
g
r
a
in
e
d
l
o
ca
l
izat
io
n
a
n
d
b
r
o
a
d
s
em
an
tic
c
o
v
e
r
a
g
e
.
Fig
u
r
e
3
.
Per
f
o
r
m
an
c
e
co
m
p
ar
is
o
n
o
f
f
ix
e
d
-
s
ize
an
d
r
ec
u
r
s
iv
e
tex
t c
h
u
n
k
in
g
s
tr
ateg
ies ac
r
o
s
s
d
if
f
er
en
t c
h
u
n
k
s
izes f
o
r
th
r
ee
r
etr
iev
al
e
v
alu
a
tio
n
m
etr
ics
Fig
u
r
e
4
r
ev
ea
ls
d
is
tin
ct
p
er
f
o
r
m
an
ce
p
atter
n
s
ac
r
o
s
s
ch
u
n
k
in
g
s
tr
ateg
ies
wh
e
n
e
v
alu
ate
d
o
n
th
r
e
e
r
etr
iev
al
m
etr
ics
an
d
o
n
e
g
en
er
atio
n
m
etr
ic.
R
ec
u
r
s
iv
e
ch
u
n
k
in
g
em
er
g
es
as
th
e
clea
r
w
in
n
er
f
o
r
r
etr
iev
al
task
s
,
ac
h
iev
in
g
co
n
s
is
ten
tly
s
u
p
er
io
r
s
co
r
es
ac
r
o
s
s
all
m
e
tr
ics,
with
p
ea
k
p
er
f
o
r
m
an
ce
at
in
ter
m
ed
iate
to
lar
g
e
ch
u
n
k
s
izes
(
7
5
0
-
1
,
2
5
0
ch
ar
ac
ter
s
)
.
T
h
is
ap
p
r
o
ac
h
s
ig
n
if
ican
tly
o
u
tp
e
r
f
o
r
m
s
b
o
th
f
ix
ed
-
s
ize
(
Fig
u
r
e
4
(
a)
)
a
n
d
s
em
an
tic
c
h
u
n
k
in
g
m
eth
o
d
s
(
Fig
u
r
e
4
(
b
)
)
.
T
h
e
L
L
M
-
b
ased
Mo
C
v
a
r
ian
t
[
2
2
]
s
h
o
ws
s
tr
o
n
g
co
m
p
etitiv
e
p
er
f
o
r
m
an
ce
,
p
ar
t
icu
lar
ly
ex
ce
llin
g
in
L
L
M
-
b
as
ed
co
n
tex
t
p
r
ec
is
io
n
(
Fig
u
r
e
4
(
c)
)
,
th
o
u
g
h
it
f
alls
ju
s
t sh
o
r
t o
f
th
e
b
est r
ec
u
r
s
iv
e
co
n
f
ig
u
r
atio
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
va
lu
a
tin
g
d
o
cu
me
n
t c
h
u
n
kin
g
a
p
p
r
o
a
ch
es fo
r
r
etri
ev
a
l
-
a
u
g
men
ted
g
e
n
era
tio
n
in
…
(
E
r
w
a
n
n
La
va
r
ec
)
1915
C
o
n
v
er
s
ely
,
s
em
an
tic
ch
u
n
k
in
g
ex
h
ib
its
a
n
o
tab
le
u
n
d
er
p
e
r
f
o
r
m
an
ce
ac
r
o
s
s
all
estab
lis
h
e
d
r
etr
iev
al
m
etr
ics.
T
h
e
b
asic
L
L
M
-
b
ased
ch
u
n
k
er
f
ar
es
s
o
m
ewh
at
b
etter
,
d
eliv
er
in
g
m
o
d
er
ate
o
u
tco
m
es
th
at
g
en
er
all
y
b
r
id
g
e
th
e
p
er
f
o
r
m
an
ce
g
ap
b
etwe
en
th
e
b
aselin
e
f
ix
e
d
-
s
ize
m
eth
o
d
s
an
d
th
e
m
o
r
e
ad
v
an
ce
d
Mo
C
ap
p
r
o
ac
h
es.
T
o
g
eth
e
r
,
th
ese
r
e
s
u
lts
h
ig
h
lig
h
t th
e
d
is
tin
ct
ad
v
an
tag
e
o
f
r
ec
u
r
s
iv
e
s
eg
m
en
tati
o
n
,
wh
ich
s
tr
ik
es a
cr
itical
b
alan
ce
b
etwe
en
p
r
eser
v
in
g
d
o
cu
m
e
n
t stru
ctu
r
e
a
n
d
m
ain
tain
in
g
s
em
an
tic
co
h
er
en
ce
.
T
h
e
g
en
er
atio
n
q
u
ality
r
esu
lts
(
Fig
u
r
e
4
(
d
)
)
r
ev
ea
l a
n
e
v
en
m
o
r
e
p
r
o
n
o
u
n
ce
d
ad
v
an
tag
e
f
o
r
r
ec
u
r
s
iv
e
ch
u
n
k
i
n
g
.
At
lar
g
er
c
h
u
n
k
s
izes
(
1
,
250
-
1
,
5
0
0
ch
a
r
ac
ter
s
)
,
r
e
cu
r
s
iv
e
s
eg
m
en
tatio
n
d
o
m
in
at
es
with
L
L
M
-
ju
d
g
e
s
co
r
es
ex
ce
ed
in
g
4
.
3
o
n
th
e
f
i
v
e
-
p
o
in
t
s
ca
le
—
r
ep
r
esen
tin
g
t
h
e
h
ig
h
est
p
e
r
f
o
r
m
an
ce
ac
r
o
s
s
all
m
eth
o
d
s
test
ed
.
W
h
ile
f
ix
ed
-
s
ize
ch
u
n
k
in
g
m
ain
tain
s
s
tead
y
p
er
f
o
r
m
an
ce
,
it
co
n
s
is
ten
tly
f
alls
s
h
o
r
t
o
f
r
ec
u
r
s
iv
e
p
ea
k
s
,
an
d
s
em
an
tic
ch
u
n
k
in
g
s
h
o
ws
m
a
r
k
ed
ly
wea
k
e
r
r
esu
lts
.
T
h
e
L
L
M
-
d
r
iv
en
a
p
p
r
o
ac
h
es
f
o
llo
w
a
f
am
iliar
p
atter
n
:
Mo
C
o
u
tp
er
f
o
r
m
s
th
e
b
asic
L
L
M
ch
u
n
k
er
b
u
t
n
eith
er
v
ar
ian
t
m
atch
es
th
e
t
o
p
-
p
e
r
f
o
r
m
in
g
r
ec
u
r
s
iv
e
co
n
f
ig
u
r
atio
n
s
.
T
h
ese
co
n
v
e
r
g
in
g
r
esu
lts
ac
r
o
s
s
b
o
th
r
etr
iev
al
an
d
g
en
er
atio
n
m
et
r
ics
h
ig
h
lig
h
t
th
e
ef
f
ec
tiv
en
ess
o
f
r
ec
u
r
s
iv
e
ch
u
n
k
in
g
,
wh
o
s
e
p
r
eser
v
ati
o
n
o
f
d
o
cu
m
e
n
t
s
tr
u
ctu
r
e
an
d
ab
ilit
y
to
ac
co
m
m
o
d
ate
m
o
d
er
ately
lar
g
e
ch
u
n
k
s
izes
p
r
o
v
id
e
an
o
p
tim
al
b
alan
ce
.
(
a)
(
b
)
(
c)
(
d
)
Fig
u
r
e
4
.
C
o
m
p
a
r
is
o
n
o
f
ch
u
n
k
in
g
s
tr
ateg
ies ac
r
o
s
s
r
etr
iev
al
m
etr
ics
of
(
a)
s
im
p
le
c
o
n
tex
t
p
r
ec
is
io
n
,
(
b
)
s
em
an
tic
s
im
ilar
ity
,
(
c)
L
L
M
-
b
ased
co
n
tex
t
p
r
ec
is
io
n
,
a
n
d
(
d
)
g
en
er
atio
n
q
u
ality
T
h
ese
f
in
d
in
g
s
h
ig
h
lig
h
t
s
ev
e
r
al
im
p
o
r
tan
t
tr
ad
e
-
o
f
f
s
f
o
r
d
esig
n
in
g
R
AG
s
y
s
tem
s
in
th
e
ed
ito
r
ial
d
o
m
ain
d
ea
lin
g
with
th
e
s
u
b
ject
o
f
SaaS.
First,
ch
u
n
k
s
ize
ex
er
ts
a
d
ir
ec
t
in
f
l
u
en
ce
o
n
r
etr
iev
al
q
u
ality
:
s
m
aller
s
eg
m
en
ts
f
ac
ilit
ate
d
o
cu
m
en
t
-
lev
el
lo
ca
lizatio
n
,
w
h
er
ea
s
m
o
d
er
ately
lar
g
er
o
n
e
s
im
p
r
o
v
e
s
em
an
tic
alig
n
m
en
t.
L
ar
g
e
r
ch
u
n
k
s
izes
g
en
er
ally
lead
to
im
p
r
o
v
ed
g
en
er
atio
n
p
er
f
o
r
m
a
n
ce
.
Ho
we
v
er
,
r
ec
en
t
f
in
d
in
g
s
f
r
o
m
C
h
r
o
m
a’
s
tech
n
ical
r
ep
o
r
t
s
h
o
w
t
h
at
wh
e
n
m
o
d
els
ar
e
ex
p
o
s
ed
to
v
e
r
y
lo
n
g
co
n
tex
t
s
,
i.e
.
,
o
n
th
e
o
r
d
e
r
o
f
1
0
4
to
k
e
n
s
,
th
eir
p
er
f
o
r
m
an
ce
ca
n
d
eg
r
a
d
e
d
u
e
to
p
o
s
itio
n
al
b
iases
an
d
r
ed
u
ce
d
r
eliab
ilit
y
,
ev
en
o
n
r
elativ
ely
s
im
p
le
task
s
[
6
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
9
0
9
-
1
9
1
8
1916
Ov
er
all,
r
ec
u
r
s
iv
e
c
h
u
n
k
in
g
e
m
er
g
es
as
th
e
m
o
s
t
ef
f
ec
tiv
e,
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
in
g
f
ix
ed
-
s
ize,
s
em
an
tic,
an
d
L
L
M
-
d
r
iv
en
a
p
p
r
o
ac
h
es
in
b
o
th
r
etr
ie
v
al
an
d
g
en
e
r
atio
n
task
s
.
Nev
er
t
h
eless
,
th
e
r
elativ
e
u
n
d
er
p
er
f
o
r
m
an
ce
o
f
s
em
an
tic
an
d
b
asic L
L
M
-
b
ased
ch
u
n
k
e
r
s
s
u
g
g
ests
th
at
s
o
p
h
is
ticated
s
eg
m
en
tatio
n
alo
n
e
d
o
es
n
o
t
g
u
ar
a
n
tee
b
etter
d
o
w
n
s
tr
ea
m
u
tili
ty
with
o
u
t
ca
r
ef
u
l
ca
lib
r
atio
n
.
W
h
ile
th
e
p
r
ese
n
t
s
tu
d
y
f
o
cu
s
es
o
n
ed
ito
r
ial
co
n
ten
t
r
eg
a
r
d
in
g
Sa
aS,
it
is
h
y
p
o
th
esized
th
at
th
e
ad
v
an
tag
es
o
f
r
ec
u
r
s
iv
e
ch
u
n
k
in
g
lik
ely
ex
ten
d
to
o
th
er
s
p
ec
ialized
an
d
n
o
n
-
s
p
ec
ialized
ed
ito
r
ial
d
o
m
ain
s
.
R
ec
en
t
r
esear
ch
in
th
e
f
ield
o
f
leg
al
d
o
cu
m
en
t
r
etr
iev
al
[
3
0
]
,
[
3
1
]
h
as
d
em
o
n
s
tr
ated
th
e
ef
f
icac
y
o
f
r
ec
u
r
s
iv
e
ch
u
n
k
in
g
in
p
r
eser
v
in
g
th
e
h
ier
ar
c
h
ical
s
tr
u
ctu
r
e
an
d
cr
o
s
s
-
r
ef
er
en
ce
s
th
at
ar
e
cr
itical
f
o
r
leg
al
in
ter
p
r
etatio
n
.
T
h
ese
f
in
d
in
g
s
s
u
g
g
est
th
at
d
o
m
ain
s
ch
ar
ac
ter
ized
b
y
in
h
er
en
t
s
tr
u
ctu
r
al
o
r
g
an
izatio
n
m
ay
p
ar
t
icu
lar
ly
b
en
e
f
it
f
r
o
m
th
is
ap
p
r
o
ac
h
.
I
n
a
s
im
ilar
v
ein
,
tech
n
ical
d
o
cu
m
en
tatio
n
an
d
s
cien
tific
p
ap
er
s
—
wh
ich
f
r
eq
u
e
n
tly
co
n
tain
s
ec
ti
o
n
s
th
at
r
ef
er
en
ce
d
ef
in
itio
n
s
o
r
co
n
ce
p
ts
in
tr
o
d
u
ce
d
elsewh
er
e
—
h
av
e
b
ee
n
d
em
o
n
s
tr
ated
to
ex
h
ib
it
o
p
tim
a
l
p
er
f
o
r
m
a
n
ce
wh
en
u
tili
zin
g
r
ec
u
r
s
iv
e
an
d
h
ier
ar
ch
ical
ch
u
n
k
in
g
s
tr
ateg
ies.
C
o
n
s
eq
u
en
tly
,
we
h
y
p
o
th
esize
th
at
th
e
ad
v
an
tag
es
o
f
r
ec
u
r
s
iv
e
c
h
u
n
k
in
g
will
tr
an
s
f
er
to
a
b
r
o
a
d
er
r
an
g
e
o
f
ed
ito
r
ial
d
o
m
ai
n
s
wh
en
r
el
iab
le
s
tr
u
ctu
r
al
o
r
f
o
r
m
attin
g
s
ig
n
als
ar
e
av
ai
lab
le.
Fu
tu
r
e
r
esear
ch
s
h
o
u
ld
em
p
ir
ically
v
alid
ate
r
e
cu
r
s
iv
e
ch
u
n
k
in
g
'
s
ef
f
ec
tiv
en
ess
ac
r
o
s
s
th
ese
d
iv
er
s
e
d
o
m
ain
s
to
estab
lis
h
its
b
r
o
ad
er
ap
p
licab
ilit
y
an
d
id
en
ti
f
y
d
o
m
ain
-
s
p
ec
if
ic
o
p
tim
izatio
n
r
eq
u
ir
em
en
ts
.
B
ey
o
n
d
cr
o
s
s
-
d
o
m
ain
ap
p
lic
ab
ilit
y
,
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
also
co
n
n
ec
t
to
b
r
o
ad
er
r
esear
ch
ef
f
o
r
ts
in
R
AG
o
p
tim
izatio
n
an
d
lo
n
g
-
co
n
tex
t
m
o
d
elin
g
.
R
ec
en
t
ev
id
en
ce
s
h
o
ws
th
at
L
L
Ms
ex
h
ib
it
co
n
te
x
t
d
ec
ay
a
n
d
f
ail
to
u
n
if
o
r
m
ly
atten
d
to
in
f
o
r
m
atio
n
d
is
tr
ib
u
ted
ac
r
o
s
s
lo
n
g
in
p
u
t
s
eq
u
en
ce
s
[
3
2
]
.
T
h
is
lim
itatio
n
h
ig
h
lig
h
ts
th
e
im
p
o
r
tan
ce
o
f
s
u
p
p
ly
in
g
m
o
d
els
with
co
h
er
en
t,
well
-
s
tr
u
ctu
r
ed
co
n
tex
t
s
eg
m
en
ts
r
ath
er
th
a
n
lo
n
g
,
n
o
is
y
co
n
ca
te
n
atio
n
s
o
f
tex
t.
B
y
d
em
o
n
s
tr
atin
g
th
a
t
r
ec
u
r
s
iv
e
ch
u
n
k
in
g
b
etter
p
r
eser
v
es
d
o
cu
m
e
n
t
s
tr
u
ctu
r
e
wh
ile
m
ain
tain
in
g
m
an
ag
ea
b
le
s
eg
m
en
t
len
g
th
s
,
o
u
r
r
esu
lts
co
n
tr
ib
u
te
to
th
is
wid
er
lin
e
o
f
r
esear
c
h
s
ee
k
in
g
to
im
p
r
o
v
e
th
e
r
o
b
u
s
tn
ess
,
f
ac
tu
al
co
n
s
is
ten
cy
,
a
n
d
ef
f
icien
cy
o
f
R
AG
s
y
s
tem
s
o
p
er
atin
g
u
n
d
er
lo
n
g
-
co
n
tex
t c
o
n
s
tr
ain
ts
.
5.
CO
NCLU
SI
O
N
T
h
is
s
t
u
d
y
s
y
s
t
em
ati
ca
ll
y
in
v
est
ig
at
ed
t
h
e
i
m
p
a
ct
o
f
d
o
cu
m
e
n
t
c
h
u
n
k
in
g
s
t
r
at
e
g
ies
o
n
R
A
G
p
e
r
f
o
r
m
a
n
c
e
wi
th
in
t
h
e
e
d
it
o
r
ial
d
o
m
a
in
r
e
g
a
r
d
in
g
Sa
a
S.
T
h
e
r
es
u
l
ts
d
e
m
o
n
s
t
r
ate
th
at
p
r
e
-
i
n
d
e
x
i
n
g
s
eg
m
e
n
t
ati
o
n
d
ec
is
io
n
s
s
t
r
o
n
g
l
y
i
n
f
lu
e
n
ce
b
o
t
h
r
et
r
i
ev
al
ef
f
e
cti
v
e
n
ess
a
n
d
an
s
we
r
q
u
a
lit
y
,
w
it
h
r
ec
u
r
s
iv
e
ch
u
n
k
in
g
u
s
i
n
g
m
o
d
e
r
at
el
y
l
ar
g
e
s
eg
m
e
n
ts
(
7
5
0
-
1
,
2
5
0
c
h
ar
ac
t
er
s
)
c
o
n
s
is
t
e
n
tl
y
y
iel
d
i
n
g
t
h
e
b
es
t
b
ala
n
ce
b
et
we
en
d
o
c
u
m
en
t
lo
ca
l
iz
ati
o
n
a
n
d
s
e
m
a
n
t
ic
c
o
h
e
r
e
n
c
e.
T
h
es
e
f
i
n
d
i
n
g
s
h
i
g
h
li
g
h
t
t
h
at
s
tr
u
ct
u
r
al
d
o
c
u
m
e
n
t
awa
r
e
n
ess
r
e
m
ai
n
s
ess
en
tia
l,
ev
en
w
h
e
n
c
o
m
p
a
r
e
d
wit
h
m
o
r
e
s
o
p
h
is
t
ica
te
d
s
e
m
a
n
ti
c
-
aw
ar
e
o
r
L
L
M
-
d
r
i
v
e
n
ch
u
n
k
in
g
m
e
th
o
d
s
,
a
n
d
th
at
ef
f
e
cti
v
e
c
h
u
n
k
i
n
g
m
u
s
t
b
ala
n
c
e
r
et
r
ie
v
al
p
r
ec
is
io
n
wit
h
to
k
e
n
b
u
d
g
e
t
a
n
d
lo
n
g
-
co
n
t
e
x
t r
e
lia
b
i
lit
y
co
n
s
t
r
a
in
ts
.
L
o
o
k
i
n
g
ah
ea
d
,
f
u
tu
r
e
r
es
ea
r
c
h
s
h
o
u
ld
g
e
n
e
r
al
iz
e
t
h
es
e
o
b
s
e
r
v
ati
o
n
s
ac
r
o
s
s
o
t
h
e
r
d
o
m
a
i
n
s
s
u
c
h
as
l
e
g
al
o
r
s
ci
e
n
ti
f
ic
te
x
ts
,
e
x
p
l
o
r
e
ad
ap
ti
v
e
ch
u
n
k
i
n
g
c
o
m
b
i
n
e
d
wit
h
h
y
b
r
i
d
s
e
ar
ch
s
tr
a
te
g
ies
m
ix
in
g
s
ev
er
al
a
p
p
r
o
ac
h
es
,
d
y
n
am
i
c
r
et
r
i
e
v
al
o
r
r
e
r
a
n
k
in
g
s
tr
ate
g
i
es,
a
n
d
i
n
v
est
ig
at
e
h
o
w
s
eg
m
e
n
t
ati
o
n
m
u
s
t
ev
o
l
v
e
al
o
n
g
s
i
d
e
em
er
g
i
n
g
lo
n
g
-
c
o
n
te
x
t
a
r
c
h
it
ec
t
u
r
es
.
B
y
p
r
o
v
id
in
g
a
d
o
m
ai
n
-
s
p
ec
i
f
i
c
b
e
n
c
h
m
a
r
k
,
a
c
o
m
p
r
eh
en
s
iv
e
ev
al
u
at
io
n
f
r
am
ew
o
r
k
,
an
d
e
m
p
i
r
ic
al
g
u
id
an
ce
o
n
o
p
ti
m
a
l
ch
u
n
k
in
g
d
esi
g
n
,
t
h
is
wo
r
k
a
d
v
a
n
c
es
b
o
t
h
t
h
e
t
h
e
o
r
e
tica
l
u
n
d
e
r
s
ta
n
d
i
n
g
a
n
d
p
r
a
cti
c
al
d
e
p
l
o
y
m
e
n
t
o
f
r
eli
ab
le
,
ef
f
i
c
ien
t RA
G
s
y
s
t
em
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
E
r
wan
n
L
av
a
r
ec
✓
✓
✓
✓
✓
✓
✓
✓
Yu
Du
✓
✓
✓
✓
✓
✓
✓
✓
✓
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 Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
va
lu
a
tin
g
d
o
cu
me
n
t c
h
u
n
kin
g
a
p
p
r
o
a
ch
es fo
r
r
etri
ev
a
l
-
a
u
g
men
ted
g
e
n
era
tio
n
in
…
(
E
r
w
a
n
n
La
va
r
ec
)
1917
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
d
ec
lar
e
th
at
th
e
y
h
av
e
n
o
co
n
f
licts
o
f
in
ter
est r
e
lated
to
th
is
wo
r
k
.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ataset
s
u
p
p
o
r
tin
g
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
is
p
u
b
licly
av
ailab
le
in
Z
en
o
d
o
at
h
ttp
s
://d
o
i.o
r
g
/1
0
.
5
2
8
1
/zen
o
d
o
.
1
7
4
2
3
3
3
7
.
T
h
e
d
ataset
is
d
is
tr
ib
u
ted
u
n
d
er
th
e
C
r
ea
tiv
e
C
o
m
m
o
n
s
Attr
ib
u
tio
n
4
.
0
I
n
te
r
n
atio
n
al
(
C
C
B
Y
4
.
0
)
licen
s
e,
wh
ich
p
er
m
its
u
n
r
es
tr
icted
u
s
e,
d
is
tr
ib
u
tio
n
,
an
d
r
ep
r
o
d
u
ctio
n
in
an
y
m
ed
iu
m
,
p
r
o
v
id
e
d
th
e
o
r
i
g
in
a
l
wo
r
k
is
p
r
o
p
er
ly
cited
.
A
m
ir
r
o
r
ed
v
er
s
io
n
o
f
th
e
d
ataset
is
also
h
o
s
ted
o
n
Hu
g
g
in
g
Face
at
h
ttp
s
://h
u
g
g
in
g
f
ac
e.
c
o
/d
atasets
/ap
p
v
izer
/q
a
-
d
ataset
-
f
o
r
-
s
aa
s
to
f
ac
ilit
ate
ac
ce
s
s
f
o
r
th
e
r
esear
ch
co
m
m
u
n
ity
.
RE
F
E
R
E
NC
E
S
[
1
]
L.
H
u
a
n
g
e
t
a
l
.
,
“
A
s
u
r
v
e
y
o
n
h
a
l
l
u
c
i
n
a
t
i
o
n
i
n
l
a
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s
:
p
r
i
n
c
i
p
l
e
s
,
t
a
x
o
n
o
my
,
c
h
a
l
l
e
n
g
e
s
,
a
n
d
o
p
e
n
q
u
e
s
t
i
o
n
s,”
AC
M
T
ra
n
s
a
c
t
i
o
n
s
o
n
I
n
f
o
rm
a
t
i
o
n
S
y
st
e
m
s
,
v
o
l
.
4
3
,
n
o
.
2
,
p
p
.
1
–
5
5
,
2
0
2
5
,
d
o
i
:
1
0
.
1
1
4
5
/
3
7
0
3
1
5
5
.
[
2
]
P
.
Le
w
i
s
e
t
a
l
.
,
“
R
e
t
r
i
e
v
a
l
-
a
u
g
me
n
t
e
d
g
e
n
e
r
a
t
i
o
n
f
o
r
k
n
o
w
l
e
d
g
e
-
i
n
t
e
n
s
i
v
e
N
LP
t
a
s
k
s,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
3
4
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
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
st
e
m
s
,
2
0
2
0
,
p
p
.
9
4
5
9
–
9
4
7
4
,
d
o
i
:
1
0
.
5
5
5
5
/
3
4
9
5
7
2
4
.
3
4
9
6
5
1
7
.
[
3
]
W
.
F
a
n
e
t
a
l
.
,
“
A
s
u
r
v
e
y
o
n
R
A
G
m
e
e
t
i
n
g
LLM
s:
t
o
w
a
r
d
s
r
e
t
r
i
e
v
a
l
-
a
u
g
m
e
n
t
e
d
l
a
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s,”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
3
0
t
h
AC
M
S
I
G
K
D
D
C
o
n
f
e
r
e
n
c
e
o
n
K
n
o
w
l
e
d
g
e
D
i
sc
o
v
e
ry
a
n
d
D
a
t
a
M
i
n
i
n
g
,
2
0
2
4
,
p
p
.
6
4
9
1
–
6
5
0
1
,
d
o
i
:
1
0
.
1
1
4
5
/
3
6
3
7
5
2
8
.
3
6
7
1
4
7
0
.
[
4
]
Y
.
G
a
o
e
t
a
l
.
,
“
R
e
t
r
i
e
v
a
l
-
a
u
g
me
n
t
e
d
g
e
n
e
r
a
t
i
o
n
f
o
r
l
a
r
g
e
l
a
n
g
u
a
g
e
mo
d
e
l
s
:
a
su
r
v
e
y
,
”
2
0
2
3
,
a
rX
i
v
:
2
3
1
2
.
1
0
9
9
7
.
[
5
]
N
.
A
mp
a
z
i
s,
“
I
mp
r
o
v
i
n
g
R
A
G
q
u
a
l
i
t
y
f
o
r
l
a
r
g
e
l
a
n
g
u
a
g
e
mo
d
e
l
s
w
i
t
h
t
o
p
i
c
-
e
n
h
a
n
c
e
d
r
e
r
a
n
k
i
n
g
,
”
i
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
Ap
p
l
i
c
a
t
i
o
n
s
a
n
d
I
n
n
o
v
a
t
i
o
n
s
,
C
h
a
m
,
S
w
i
t
z
e
r
l
a
n
d
:
S
p
r
i
n
g
e
r
,
2
0
2
4
,
p
p
.
7
4
–
8
7
.
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
1
-
6
3
2
1
5
-
0
_
6
.
[
6
]
K
.
H
o
n
g
,
A
.
Tr
o
y
n
i
k
o
v
,
a
n
d
J.
H
u
b
e
r
,
“
C
o
n
t
e
x
t
r
o
t
:
h
o
w
i
n
c
r
e
a
si
n
g
i
n
p
u
t
t
o
k
e
n
s
i
m
p
a
c
t
s
L
LM
p
e
r
f
o
r
ma
n
c
e
,
”
C
h
ro
m
a
R
e
se
a
rc
h
.
2
0
2
5
,
A
c
c
e
s
se
d
:
S
e
p
.
0
2
,
2
0
2
5
,
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s:
/
/
r
e
s
e
a
r
c
h
.
t
r
y
c
h
r
o
ma
.
c
o
m/
c
o
n
t
e
x
t
-
r
o
t
[
7
]
S
.
W
a
n
g
,
J.
Ta
n
,
Z.
D
o
u
,
a
n
d
J.
-
R
.
W
e
n
,
“
O
m
n
i
Ev
a
l
:
a
n
o
m
n
i
d
i
r
e
c
t
i
o
n
a
l
a
n
d
a
u
t
o
ma
t
i
c
R
A
G
e
v
a
l
u
a
t
i
o
n
b
e
n
c
h
mar
k
i
n
f
i
n
a
n
c
i
a
l
d
o
m
a
i
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
2
5
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
,
2
0
2
5
,
p
p
.
5
7
3
7
–
5
7
6
2
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
5
.
e
m
n
l
p
-
m
a
i
n
.
2
9
2
.
[
8
]
K
.
Z
h
u
e
t
a
l
.
,
“
R
A
G
Ev
a
l
:
s
c
e
n
a
r
i
o
s
p
e
c
i
f
i
c
R
A
G
e
v
a
l
u
a
t
i
o
n
d
a
t
a
set
g
e
n
e
r
a
t
i
o
n
f
r
a
mew
o
r
k
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
6
3
rd
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
,
2
0
2
5
,
p
p
.
8
5
2
0
–
8
5
4
4
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
5
.
a
c
l
-
l
o
n
g
.
4
1
8
.
[
9
]
C
.
M
e
r
o
l
a
a
n
d
J.
S
i
n
g
h
,
“
R
e
c
o
n
st
r
u
c
t
i
n
g
c
o
n
t
e
x
t
:
e
v
a
l
u
a
t
i
n
g
a
d
v
a
n
c
e
d
c
h
u
n
k
i
n
g
st
r
a
t
e
g
i
e
s
f
o
r
r
e
t
r
i
e
v
a
l
-
a
u
g
m
e
n
t
e
d
g
e
n
e
r
a
t
i
o
n
,
”
i
n
K
n
o
w
l
e
d
g
e
-
E
n
h
a
n
c
e
d
I
n
f
o
rm
a
t
i
o
n
Re
t
ri
e
v
a
l
,
C
h
a
m
,
S
w
i
t
z
e
r
l
a
n
d
:
S
p
r
i
n
g
e
r
,
2
0
2
6
,
p
p
.
3
–
18
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
2
-
0
2
8
9
9
-
0
_
1
.
[
1
0
]
L.
Z
h
e
n
g
e
t
a
l
.
,
“
J
u
d
g
i
n
g
L
LM
-
as
-
a
-
j
u
d
g
e
w
i
t
h
M
T
-
b
e
n
c
h
a
n
d
c
h
a
t
b
o
t
a
r
e
n
a
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
3
7
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
st
e
m
s
,
2
0
2
3
,
v
o
l
.
3
6
,
p
p
.
4
6
5
9
5
–
4
6
6
2
3
,
d
o
i
:
1
0
.
5
5
5
5
/
3
6
6
6
1
2
2
.
3
6
6
8
1
4
2
.
[
1
1
]
Y
.
A
.
M
a
l
k
o
v
a
n
d
D
.
A
.
Y
a
s
h
u
n
i
n
,
“
Ef
f
i
c
i
e
n
t
a
n
d
r
o
b
u
s
t
a
p
p
r
o
x
i
ma
t
e
n
e
a
r
e
st
n
e
i
g
h
b
o
r
se
a
r
c
h
u
si
n
g
h
i
e
r
a
r
c
h
i
c
a
l
n
a
v
i
g
a
b
l
e
sma
l
l
w
o
r
l
d
g
r
a
p
h
s,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
A
n
a
l
y
s
i
s
a
n
d
Ma
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
4
2
,
n
o
.
4
,
p
p
.
8
2
4
–
8
3
6
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
TPA
M
I
.
2
0
1
8
.
2
8
8
9
4
7
3
.
[
1
2
]
H
.
Jé
g
o
u
,
M
.
D
o
u
z
e
,
a
n
d
C
.
S
c
h
m
i
d
,
“
P
r
o
d
u
c
t
q
u
a
n
t
i
z
a
t
i
o
n
f
o
r
n
e
a
r
e
st
n
e
i
g
h
b
o
r
se
a
r
c
h
,
”
I
EE
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
Pa
t
t
e
r
n
A
n
a
l
y
si
s
a
n
d
Ma
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
3
3
,
n
o
.
1
,
p
p
.
1
1
7
–
1
2
8
,
2
0
1
1
,
d
o
i
:
1
0
.
1
1
0
9
/
T
P
A
M
I
.
2
0
1
0
.
5
7
.
[
1
3
]
A
.
G
i
o
n
i
s,
P
.
I
n
d
y
k
,
a
n
d
R
.
M
o
t
w
a
n
i
,
“
S
i
mi
l
a
r
i
t
y
se
a
r
c
h
i
n
h
i
g
h
d
i
me
n
s
i
o
n
s
v
i
a
h
a
s
h
i
n
g
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
5
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Ve
r
y
L
a
rg
e
D
a
t
a
B
a
s
e
s
,
1
9
9
9
,
p
p
.
5
1
8
–
529
,
d
o
i
:
1
0
.
5
5
5
5
/
6
4
5
9
2
5
.
6
7
1
5
1
6
.
[
1
4
]
Z.
X
u
e
t
a
l
.
,
“
R
e
t
r
i
e
v
a
l
-
a
u
g
me
n
t
e
d
g
e
n
e
r
a
t
i
o
n
w
i
t
h
k
n
o
w
l
e
d
g
e
g
r
a
p
h
s
f
o
r
c
u
s
t
o
m
e
r
ser
v
i
c
e
q
u
e
s
t
i
o
n
a
n
sw
e
r
i
n
g
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
4
7
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
A
C
M
S
I
G
I
R
C
o
n
f
e
re
n
c
e
o
n
Re
s
e
a
r
c
h
a
n
d
D
e
v
e
l
o
p
m
e
n
t
i
n
I
n
f
o
rm
a
t
i
o
n
R
e
t
r
i
e
v
a
l
,
2
0
2
4
,
p
p
.
2
9
0
5
–
2
9
0
9
,
d
o
i
:
1
0
.
1
1
4
5
/
3
6
2
6
7
7
2
.
3
6
6
1
3
7
0
.
[
1
5
]
S
.
Li
,
L.
S
t
e
n
z
e
l
,
C
.
E
i
c
k
h
o
f
f
,
a
n
d
S
.
A
l
i
B
a
h
r
a
i
n
i
a
n
,
“
En
h
a
n
c
i
n
g
r
e
t
r
i
e
v
a
l
-
a
u
g
me
n
t
e
d
g
e
n
e
r
a
t
i
o
n
:
a
s
t
u
d
y
o
f
b
e
st
p
r
a
c
t
i
c
e
s,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
3
1
st
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
st
i
c
s
,
2
0
2
5
,
p
p
.
6
7
0
5
–
6
7
1
7
.
[
1
6
]
Y
.
Y
u
e
t
a
l
.
,
“
R
a
n
k
R
A
G
:
u
n
i
f
y
i
n
g
c
o
n
t
e
x
t
r
a
n
k
i
n
g
w
i
t
h
r
e
t
r
i
e
v
a
l
-
a
u
g
me
n
t
e
d
g
e
n
e
r
a
t
i
o
n
i
n
LL
M
s,”
i
n
3
8
t
h
C
o
n
f
e
ren
c
e
o
n
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
st
e
m
s (N
e
u
rI
PS
2
0
2
4
)
,
2
0
2
4
,
p
p
.
1
2
1
1
5
6
–
1
2
1
1
8
4
.
d
o
i
:
1
0
.
5
2
2
0
2
/
0
7
9
0
1
7
-
3
8
5
0
.
[
1
7
]
L.
G
a
o
,
X
.
M
a
,
J.
Li
n
,
a
n
d
J.
C
a
l
l
a
n
,
“
P
r
e
c
i
se
z
e
r
o
-
sh
o
t
d
e
n
se
r
e
t
r
i
e
v
a
l
w
i
t
h
o
u
t
r
e
l
e
v
a
n
c
e
l
a
b
e
l
s,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
6
1
st
An
n
u
a
l
M
e
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
,
2
0
2
3
,
p
p
.
1
7
6
2
–
1
7
7
7
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
3
.
a
c
l
-
l
o
n
g
.
9
9
.
[
1
8
]
M
.
A
.
C
.
B
l
a
n
d
ó
n
,
J.
T
a
l
u
r
,
B
.
C
h
a
r
r
o
n
,
D
.
L
i
u
,
S
.
M
a
n
s
o
u
r
,
a
n
d
M
.
F
e
d
e
r
i
c
o
,
“
M
E
M
E
R
A
G
:
a
m
u
l
t
i
l
i
n
g
u
a
l
e
n
d
-
to
-
e
n
d
met
a
-
e
v
a
l
u
a
t
i
o
n
b
e
n
c
h
m
a
r
k
f
o
r
r
e
t
r
i
e
v
a
l
a
u
g
m
e
n
t
e
d
g
e
n
e
r
a
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
6
3
r
d
A
n
n
u
a
l
M
e
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
s
t
i
c
s
,
2
0
2
5
,
p
p
.
2
2
5
7
7
–
2
2
5
9
5
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
5
.
a
c
l
-
l
o
n
g
.
1
1
0
1
.
[
1
9
]
La
n
g
C
h
a
i
n
,
“
L
a
n
g
C
h
a
i
n
o
v
e
r
v
i
e
w
,”
L
a
n
g
C
h
a
i
n
D
o
c
u
m
e
n
t
s
,
A
c
c
e
ss
e
d
:
S
e
p
.
0
2
,
2
0
2
5
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
p
y
t
h
o
n
.
l
a
n
g
c
h
a
i
n
.
c
o
m
/
d
o
c
s/
h
o
w
_
t
o
/
s
e
ma
n
t
i
c
-
c
h
u
n
k
e
r
/
[
2
0
]
Ll
a
m
a
I
n
d
e
x
,
“
S
e
ma
n
t
i
c
c
h
u
n
k
e
r
,
”
L
l
a
m
a
I
n
d
e
x
,
2
0
2
4
,
A
c
c
e
s
sed
:
S
e
p
.
0
2
,
2
0
2
5
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
d
e
v
e
l
o
p
e
r
s.
l
l
a
ma
i
n
d
e
x
.
a
i
/
p
y
t
h
o
n
/
e
x
a
mp
l
e
s
/
n
o
d
e
_
p
a
r
sers
/
sem
a
n
t
i
c
_
c
h
u
n
k
i
n
g
/
[
2
1
]
R
.
Q
u
,
R
.
T
u
,
a
n
d
F
.
S
.
B
a
o
,
“
I
s
sem
a
n
t
i
c
c
h
u
n
k
i
n
g
w
o
r
t
h
t
h
e
c
o
m
p
u
t
a
t
i
o
n
a
l
c
o
s
t
?
”
i
n
F
i
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
s
t
i
c
s:
N
AAC
L
2
0
2
5
,
2
0
2
5
,
p
p
.
2
1
5
5
–
2
1
7
7
,
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
.
1
1
4
.
[
2
2
]
J.
Z
h
a
o
e
t
a
l
.
,
“
M
o
C
:
m
i
x
t
u
r
e
s
o
f
t
e
x
t
c
h
u
n
k
i
n
g
l
e
a
r
n
e
r
s
f
o
r
r
e
t
r
i
e
v
a
l
-
a
u
g
men
t
e
d
g
e
n
e
r
a
t
i
o
n
sy
s
t
e
m
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
6
3
r
d
An
n
u
a
l
M
e
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
,
2
0
2
5
,
p
p
.
5
1
7
2
–
5
1
8
9
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
5
.
a
c
l
-
l
o
n
g
.
2
5
8
.
[
2
3
]
A
.
V
.
D
u
a
r
t
e
,
J
.
D
.
M
a
r
q
u
e
s
,
M
.
G
r
a
ç
a
,
M
.
F
r
e
i
r
e
,
L.
Li
,
a
n
d
A
.
L.
O
l
i
v
e
i
r
a
,
“
L
u
m
b
e
r
C
h
u
n
k
e
r
:
l
o
n
g
-
f
o
r
m
n
a
r
r
a
t
i
v
e
d
o
c
u
me
n
t
seg
m
e
n
t
a
t
i
o
n
,
”
i
n
Fi
n
d
i
n
g
s
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:
E
MN
L
P
2
0
2
4
,
2
0
2
4
,
p
p
.
6
4
7
3
–
6
4
8
6
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
4
.
f
i
n
d
i
n
g
s
-
e
mn
l
p
.
3
7
7
.
[
2
4
]
I
.
S
.
S
i
n
g
h
e
t
a
l
.
,
“
C
h
u
n
k
R
A
G
:
n
o
v
e
l
LLM
-
c
h
u
n
k
f
i
l
t
e
r
i
n
g
m
e
t
h
o
d
f
o
r
R
A
G
sy
st
e
ms,
”
2
0
2
4
,
a
rX
i
v
:
2
4
1
0
.
1
9
5
7
2
.
[
2
5
]
M
.
G
ü
n
t
h
e
r
,
I
.
M
o
h
r
,
D
.
J.
W
i
l
l
i
a
ms,
B
.
W
a
n
g
,
a
n
d
H
.
X
i
a
o
,
“
La
t
e
c
h
u
n
k
i
n
g
:
c
o
n
t
e
x
t
u
a
l
c
h
u
n
k
e
m
b
e
d
d
i
n
g
s
u
s
i
n
g
l
o
n
g
-
c
o
n
t
e
x
t
e
mb
e
d
d
i
n
g
m
o
d
e
l
s
,
”
2
0
2
5
,
a
rXi
v
:
2
4
0
9
.
0
4
7
0
1
.
[
2
6
]
A
n
t
h
r
o
p
i
c
,
“
I
n
t
r
o
d
u
c
i
n
g
c
o
n
t
e
x
t
u
a
l
r
e
t
r
i
e
v
a
l
,
”
A
n
t
h
r
o
p
i
c
.
2
0
2
4
,
A
c
c
e
ss
e
d
:
S
e
p
.
0
2
,
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
.
a
n
t
h
r
o
p
i
c
.
c
o
m
/
n
e
w
s/
c
o
n
t
e
x
t
u
a
l
-
r
e
t
r
i
e
v
a
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
9
0
9
-
1
9
1
8
1918
[
2
7
]
S
.
F
i
l
i
c
e
,
G
.
H
o
r
o
w
i
t
z
,
D
.
C
a
r
mel
,
Z.
K
a
r
n
i
n
,
L.
L
.
-
E
y
t
a
n
,
a
n
d
Y
.
M
a
a
r
e
k
,
“
G
e
n
e
r
a
t
i
n
g
Q
&A
b
e
n
c
h
m
a
r
k
s
f
o
r
R
A
G
e
v
a
l
u
a
t
i
o
n
i
n
e
n
t
e
r
p
r
i
se
set
t
i
n
g
s,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
6
3
r
d
A
n
n
u
a
l
M
e
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
,
2
0
2
5
,
p
p
.
4
6
9
–
484
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
5
.
a
c
l
-
i
n
d
u
s
t
r
y
.
3
3
.
[
2
8
]
S
.
Es
,
J
.
J
a
mes
,
L
.
E
.
A
n
k
e
,
a
n
d
S
.
S
c
h
o
c
k
a
e
r
t
,
“
R
A
G
A
s:
a
u
t
o
ma
t
e
d
e
v
a
l
u
a
t
i
o
n
o
f
r
e
t
r
i
e
v
a
l
a
u
g
me
n
t
e
d
g
e
n
e
r
a
t
i
o
n
,
”
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
E
u
r
o
p
e
a
n
C
h
a
p
t
e
r
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:
S
y
st
e
m
D
e
m
o
n
s
t
r
a
t
i
o
n
s
,
2
0
2
4
,
p
p
.
1
5
0
–
158
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
4
.
e
a
c
l
-
d
e
mo
.
1
6
.
[
2
9
]
J.
W
e
i
e
t
a
l
.
,
“
C
h
a
i
n
-
of
-
t
h
o
u
g
h
t
p
r
o
mp
t
i
n
g
e
l
i
c
i
t
s
r
e
a
so
n
i
n
g
i
n
l
a
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s,”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
3
6
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
st
e
m
s
,
2
0
2
2
,
p
p
.
2
4
8
2
4
–
2
4
8
3
7
,
d
o
i
:
1
0
.
5
5
5
5
/
3
6
0
0
2
7
0
.
3
6
0
2
0
7
0
.
[
3
0
]
A
.
F
.
F
e
r
r
a
r
i
s,
D
.
A
u
d
r
i
t
o
,
G
.
S
i
r
a
g
u
s
a
,
a
n
d
A
.
P
i
o
v
a
n
o
,
“
Le
g
a
l
c
h
u
n
k
i
n
g
:
e
v
a
l
u
a
t
i
n
g
me
t
h
o
d
s
f
o
r
e
f
f
e
c
t
i
v
e
l
e
g
a
l
t
e
x
t
r
e
t
r
i
e
v
a
l
,
”
i
n
L
e
g
a
l
K
n
o
w
l
e
d
g
e
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
y
st
e
m
,
I
O
S
P
r
e
ss,
2
0
2
4
,
p
p
.
2
7
5
–
2
8
1
,
d
o
i
:
1
0
.
3
2
3
3
/
F
A
I
A
2
4
1
2
5
5
.
[
3
1
]
N
.
P
i
p
i
t
o
n
e
a
n
d
G
.
H
.
A
l
a
mi
,
“
Le
g
a
l
B
e
n
c
h
-
R
A
G
:
a
b
e
n
c
h
mar
k
f
o
r
r
e
t
r
i
e
v
a
l
-
a
u
g
me
n
t
e
d
g
e
n
e
r
a
t
i
o
n
i
n
t
h
e
l
e
g
a
l
d
o
ma
i
n
,
”
2
0
2
4
,
a
rXi
v
:
2
4
0
8
.
1
0
3
4
3
.
[
3
2
]
N
.
F
.
Li
u
e
t
a
l
.
,
“
L
o
st
i
n
t
h
e
m
i
d
d
l
e
:
h
o
w
l
a
n
g
u
a
g
e
mo
d
e
l
s u
se
l
o
n
g
c
o
n
t
e
x
t
s,”
T
r
a
n
sa
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
2
,
p
p
.
1
5
7
–
1
7
3
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
6
2
/
t
a
c
l
_
a
_
0
0
6
3
8
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Er
wa
n
n
La
v
a
r
e
c
h
o
ld
s
a
P
h
.
D.
i
n
r
o
b
o
ti
c
s
fro
m
LIRM
M
,
Un
i
v
e
rsity
o
f
M
o
n
t
p
e
ll
ier
(
2
0
0
1
),
w
h
e
re
h
is
d
i
ss
e
rtatio
n
fo
c
u
se
d
o
n
3
D
m
o
t
io
n
e
stim
a
ti
o
n
u
si
n
g
a
c
a
m
e
ra
a
n
d
p
r
o
p
r
io
c
e
p
ti
v
e
se
n
so
rs.
He
a
lso
e
a
rn
e
d
a
n
e
x
e
c
u
ti
v
e
M
BA
fro
m
M
o
n
tp
e
ll
ier
Bu
si
n
e
ss
S
c
h
o
o
l
(2
0
1
3
)
.
He
is
c
u
rre
n
tl
y
c
h
ief
tec
h
n
o
l
o
g
y
o
ffice
r
a
t
Ap
p
v
ize
r
(S
o
fto
n
ic
g
r
o
u
p
)
a
n
d
p
re
v
io
u
sly
f
o
u
n
d
e
d
a
n
d
led
Wan
y
Ro
b
o
ti
c
s.
His
in
tere
sts
sp
a
n
m
o
b
il
e
-
ro
b
o
t
l
o
c
a
li
z
a
ti
o
n
,
p
a
rti
c
le
fil
ters
/M
o
n
te
-
Ca
rlo
m
e
th
o
d
s,
3
D
v
isio
n
,
a
n
d
t
h
e
e
n
g
i
n
e
e
rin
g
o
f
LL
M
s
a
n
d
a
g
e
n
t
ic
AI
sy
ste
m
s
su
c
h
a
s
RAG
,
se
m
a
n
ti
c
re
a
so
n
in
g
,
a
n
d
m
u
lt
i
-
a
g
e
n
t
a
rc
h
it
e
c
tu
re
.
He
is
a
lso
li
ste
d
a
s
a
n
in
v
e
n
to
r
o
n
r
o
b
o
ti
c
s
a
n
d
AI
p
a
ten
ts.
He
wo
n
th
e
n
a
ti
o
n
a
l
c
o
m
p
e
ti
ti
o
n
f
o
r
in
n
o
v
a
ti
v
e
tec
h
n
o
l
o
g
y
c
o
m
p
a
n
y
c
re
a
ti
o
n
o
rg
a
n
ize
d
b
y
t
h
e
F
re
n
c
h
M
in
istr
y
o
f
Re
se
a
rc
h
in
1
9
9
9
a
n
d
2
0
0
0
.
He
h
a
s
p
u
b
li
sh
e
d
re
se
a
rc
h
p
a
p
e
rs
in
i
n
tern
a
ti
o
n
a
l
c
o
n
fe
re
n
c
e
s
i
n
c
lu
d
in
g
IE
EE
ICRA
,
on
k
in
e
m
a
ti
c
a
n
d
d
y
n
a
m
ic m
o
d
e
li
n
g
o
f
m
u
lt
i
-
r
o
to
r
a
ircra
ft
sy
ste
m
s.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
e
rwa
n
n
.
lav
a
re
c
@a
p
p
v
ize
r.
c
o
m
.
Yu
Du
re
c
e
iv
e
d
h
is
P
h
.
D.
i
n
Co
m
p
u
ter
S
c
ien
c
e
fr
o
m
IM
T
M
i
n
e
s
Alè
s
in
2
0
2
1
with
t
h
e
d
isse
rtatio
n
“
De
s
d
o
n
n
é
e
s
a
u
x
c
o
n
n
a
issa
n
c
e
s:
v
e
rs
d
e
s
re
c
o
m
m
a
n
d
a
ti
o
n
s
p
l
u
s
p
e
rti
n
e
n
tes
,
d
i
v
e
rsifiée
s
e
t
tran
sp
a
re
n
tes
.
”
He
a
lso
h
o
l
d
s
a
n
M
.
S
c
.
in
Co
m
p
u
ter
S
c
ie
n
c
e
(
d
a
ta,
k
n
o
wle
d
g
e
a
n
d
n
a
t
u
ra
l
lan
g
u
a
g
e
p
ro
c
e
ss
in
g
)
fr
o
m
th
e
Un
iv
e
rsity
o
f
M
o
n
t
p
e
ll
ier
(2
0
1
7
)
a
n
d
a
B.
S
c
.
in
Co
m
p
u
ter
S
c
ien
c
e
fro
m
Un
iv
e
rsité
Clerm
o
n
t
Au
v
e
rg
n
e
(2
0
1
4
).
He
is
c
u
rre
n
tl
y
a
n
AI
re
se
a
rc
h
e
n
g
in
e
e
r
a
n
d
d
a
ta
sc
ien
ti
st
a
t
Ap
p
v
ize
r
(S
o
ft
o
n
ic
g
ro
u
p
).
His
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
m
a
c
h
in
e
lea
rn
in
g
,
re
c
o
m
m
e
n
d
e
r
sy
ste
m
s,
k
n
o
wle
d
g
e
e
n
g
i
n
e
e
rin
g
,
k
n
o
wle
d
g
e
g
ra
p
h
s,
n
a
tu
ra
l
lan
g
u
a
g
e
p
ro
c
e
ss
in
g
,
in
fo
rm
a
ti
o
n
re
tri
e
v
a
l,
a
n
d
a
g
e
n
ti
c
AI.
P
rio
r
t
o
h
is
c
u
rre
n
t
p
o
siti
o
n
,
h
e
wo
r
k
e
d
a
s
a
re
se
a
rc
h
e
n
g
i
n
e
e
r
a
t
Cirad
,
d
e
v
e
lo
p
in
g
se
m
a
n
ti
c
we
b
to
o
ls
fo
r
in
fo
rm
a
ti
o
n
re
tri
e
v
a
l
a
n
d
d
a
ta
a
n
a
ly
sis.
He
h
a
s
p
u
b
li
s
h
e
d
se
v
e
ra
l
p
a
p
e
rs
in
i
n
tern
a
ti
o
n
a
l
jo
u
r
n
a
ls
in
c
lu
d
i
n
g
I
n
fo
rm
a
ti
o
n
P
r
o
c
e
ss
in
g
a
n
d
M
a
n
a
g
e
m
e
n
t,
K
n
o
wle
d
g
e
-
Ba
se
d
S
y
ste
m
s,
a
n
d
P
Lo
S
ON
E,
with
n
o
tab
le
c
o
n
tri
b
u
ti
o
n
s
to
c
o
l
lab
o
ra
ti
v
e
fi
lt
e
rin
g
,
re
c
o
m
m
e
n
d
a
ti
o
n
d
iv
e
rsit
y
,
a
n
d
e
x
p
lain
a
b
le
AI.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
d
u
.
y
u
.
1
4
1
1
@
g
m
a
il
.
c
o
m
o
r
y
u
.
d
u
@a
p
p
v
ize
r.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.