-DOCSTART- NN O O

English NN O I-Material
is NN O O
shown NN O O
to NN O O
be NN O O
trans-context-free NN O O
on NN O O
the NN O O
basis NN O O
of NN O O
coordinations NN O I-OtherScientificTerm
of NN O O
the NN O O
respectively NN O O
type NN O O
that NN O O
involve NN O O
strictly NN O I-OtherScientificTerm
syntactic NN O I-OtherScientificTerm
cross-serial NN O I-OtherScientificTerm
agreement NN O I-OtherScientificTerm
. NN O O

The NN O O
agreement NN O I-Generic
in NN O O
question NN O O
involves NN O O
number NN O O
in NN O O
nouns NN O I-OtherScientificTerm
and NN O O
reflexive NN O I-OtherScientificTerm
pronouns NN O I-OtherScientificTerm
and NN O O
is NN O O
syntactic NN O O
rather NN O O
than NN O O
semantic NN O O
in NN O O
nature NN O O
because NN O O
grammatical NN O I-OtherScientificTerm
number NN O I-OtherScientificTerm
in NN O O
English NN O I-Material
, NN O O
like NN O O
grammatical NN O I-OtherScientificTerm
gender NN O I-OtherScientificTerm
in NN O O
languages NN O I-Material
such NN O O
as NN O O
French NN O I-Material
, NN O O
is NN O O
partly NN O O
arbitrary NN O O
. NN O O

The NN O O
formal NN O O
proof NN O O
, NN O O
which NN O O
makes NN O O
crucial NN O O
use NN O O
of NN O O
the NN O O
Interchange NN O I-Method
Lemma NN O I-Method
of NN O O
Ogden NN O O
et NN O O
al. NN O O
, NN O O
is NN O O
so NN O O
constructed NN O O
as NN O O
to NN O O
be NN O O
valid NN O O
even NN O O
if NN O O
English NN O I-Material
is NN O O
presumed NN O O
to NN O O
contain NN O O
grammatical NN O O
sentences NN O O
in NN O O
which NN O O
respectively NN O O
operates NN O O
across NN O O
a NN O O
pair NN O O
of NN O O
coordinate NN O O
phrases NN O O
one NN O O
of NN O O
whose NN O O
members NN O O
has NN O O
fewer NN O O
conjuncts NN O O
than NN O O
the NN O O
other NN O O
; NN O O
it NN O O
thus NN O O
goes NN O O
through NN O O
whatever NN O O
the NN O O
facts NN O O
may NN O O
be NN O O
regarding NN O O
constructions NN O O
with NN O O
unequal NN O O
numbers NN O O
of NN O O
conjuncts NN O O
in NN O O
the NN O O
scope NN O O
of NN O O
respectively NN O O
, NN O O
whereas NN O O
other NN O O
arguments NN O O
have NN O O
foundered NN O O
on NN O O
this NN O O
problem NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
a NN O O
novel NN O O
method NN O I-Method
to NN O O
learn NN O O
the NN O O
intrinsic NN O I-OtherScientificTerm
object NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
for NN O O
robust NN O I-Task
visual NN O I-Task
tracking NN O I-Task
is NN O O
proposed NN O O
. NN O O

The NN O O
basic NN O O
assumption NN O O
is NN O O
that NN O O
the NN O O
parameterized NN O I-OtherScientificTerm
object NN O I-OtherScientificTerm
state NN O I-OtherScientificTerm
lies NN O O
on NN O O
a NN O O
low NN O I-OtherScientificTerm
dimensional NN O I-OtherScientificTerm
manifold NN O I-OtherScientificTerm
and NN O O
can NN O O
be NN O O
learned NN O O
from NN O O
training NN O O
data NN O O
. NN O O

Based NN O O
on NN O O
this NN O O
assumption NN O O
, NN O O
firstly NN O O
we NN O O
derived NN O O
the NN O O
dimensionality NN O I-Method
reduction NN O I-Method
and NN O I-Method
density NN O I-Method
estimation NN O I-Method
algorithm NN O I-Method
for NN O O
unsupervised NN O I-Task
learning NN O I-Task
of NN O I-Task
object NN O I-Method
intrinsic NN O I-Method
representation NN O I-Method
, NN O O
the NN O O
obtained NN O O
non-rigid NN O I-OtherScientificTerm
part NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
object NN O I-OtherScientificTerm
state NN O I-OtherScientificTerm
reduces NN O O
even NN O O
to NN O O
2 NN O O
dimensions NN O O
. NN O O

Secondly NN O O
the NN O O
dynamical NN O I-Method
model NN O I-Method
is NN O O
derived NN O O
and NN O O
trained NN O O
based NN O O
on NN O O
this NN O O
intrinsic NN O I-Method
representation NN O I-Method
. NN O O

Thirdly NN O O
the NN O O
learned NN O O
intrinsic NN O I-OtherScientificTerm
object NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
is NN O O
integrated NN O O
into NN O O
a NN O O
particle-filter NN O I-Method
style NN O I-Method
tracker NN O I-Method
. NN O O

We NN O O
will NN O O
show NN O O
that NN O O
this NN O O
intrinsic NN O I-Method
object NN O I-Method
representation NN O I-Method
has NN O O
some NN O O
interesting NN O O
properties NN O O
and NN O O
based NN O O
on NN O O
which NN O O
the NN O O
newly NN O O
derived NN O O
dynamical NN O I-Method
model NN O I-Method
makes NN O O
particle-filter NN O I-Method
style NN O I-Method
tracker NN O I-Method
more NN O O
robust NN O O
and NN O O
reliable NN O O
. NN O O

Experiments NN O O
show NN O O
that NN O O
the NN O O
learned NN O O
tracker NN O I-Generic
performs NN O O
much NN O O
better NN O O
than NN O O
existing NN O O
trackers NN O I-Generic
on NN O O
the NN O O
tracking NN O I-Task
of NN O I-Task
complex NN O I-OtherScientificTerm
non-rigid NN O I-OtherScientificTerm
motions NN O I-OtherScientificTerm
such NN O O
as NN O O
fish NN O I-OtherScientificTerm
twisting NN O I-OtherScientificTerm
with NN O O
self-occlusion NN O I-OtherScientificTerm
and NN O O
large NN O O
inter-frame NN O I-OtherScientificTerm
lip NN O I-OtherScientificTerm
motion NN O I-OtherScientificTerm
. NN O O

The NN O O
proposed NN O O
method NN O I-Generic
also NN O O
has NN O O
the NN O O
potential NN O O
to NN O O
solve NN O O
other NN O O
type NN O O
of NN O O
tracking NN O I-Task
problems NN O I-Task
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
a NN O O
digital NN O I-Method
signal NN O I-Method
processor NN O I-Method
-LRB- NN O I-Method
DSP NN O I-Method
-RRB- NN O I-Method
implementation NN O I-Method
of NN O O
real-time NN O I-Method
statistical NN O I-Method
voice NN O I-Method
conversion NN O I-Method
-LRB- NN O I-Method
VC NN O I-Method
-RRB- NN O I-Method
for NN O O
silent NN O I-Task
speech NN O I-Task
enhancement NN O I-Task
and NN O O
electrolaryngeal NN O I-Task
speech NN O I-Task
enhancement NN O I-Task
. NN O O

As NN O O
a NN O O
silent NN O I-OtherScientificTerm
speech NN O I-OtherScientificTerm
interface NN O I-OtherScientificTerm
, NN O O
we NN O O
focus NN O O
on NN O O
non-audible NN O I-Material
murmur NN O I-Material
-LRB- NN O I-Material
NAM NN O I-Material
-RRB- NN O I-Material
, NN O O
which NN O O
can NN O O
be NN O O
used NN O O
in NN O O
situations NN O O
where NN O O
audible NN O I-Material
speech NN O I-Material
is NN O O
not NN O O
acceptable NN O O
. NN O O

Electrolaryngeal NN O I-Material
speech NN O I-Material
is NN O O
one NN O O
of NN O O
the NN O O
typical NN O O
types NN O O
of NN O O
alaryngeal NN O I-Material
speech NN O I-Material
produced NN O O
by NN O O
an NN O O
alternative NN O O
speaking NN O I-Method
method NN O I-Method
for NN O O
laryngectomees NN O I-OtherScientificTerm
. NN O O

However NN O O
, NN O O
the NN O O
sound NN O I-Metric
quality NN O I-Metric
of NN O O
NAM NN O I-Material
and NN O I-Material
electrolaryngeal NN O I-Material
speech NN O I-Material
suffers NN O O
from NN O O
lack NN O O
of NN O O
naturalness NN O O
. NN O O

VC NN O I-Method
has NN O O
proven NN O O
to NN O O
be NN O O
one NN O O
of NN O O
the NN O O
promising NN O O
approaches NN O O
to NN O O
address NN O O
this NN O O
problem NN O O
, NN O O
and NN O O
it NN O I-Generic
has NN O O
been NN O O
successfully NN O O
implemented NN O O
on NN O O
devices NN O I-Generic
with NN O O
sufficient NN O I-Material
computational NN O I-Material
resources NN O I-Material
. NN O O

An NN O O
implementation NN O O
on NN O O
devices NN O I-Generic
that NN O O
are NN O O
highly NN O O
portable NN O O
but NN O O
have NN O O
limited NN O I-Material
computational NN O I-Material
resources NN O I-Material
would NN O O
greatly NN O O
contribute NN O O
to NN O O
its NN O O
practical NN O O
use NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
further NN O O
implement NN O O
real-time NN O I-Method
VC NN O I-Method
on NN O O
a NN O O
DSP NN O I-OtherScientificTerm
. NN O O

To NN O O
implement NN O O
the NN O O
two NN O O
speech NN O I-Method
enhancement NN O I-Method
systems NN O I-Method
based NN O O
on NN O O
real-time NN O I-Method
VC NN O I-Method
, NN O O
one NN O I-Generic
from NN O O
NAM NN O I-Material
to NN O O
a NN O O
whispered NN O I-Material
voice NN O I-Material
and NN O O
the NN O O
other NN O I-Generic
from NN O O
electrolaryngeal NN O I-Material
speech NN O I-Material
to NN O O
a NN O O
natural NN O I-Material
voice NN O I-Material
, NN O O
we NN O O
propose NN O O
several NN O O
methods NN O I-Generic
for NN O O
reducing NN O O
computational NN O I-Metric
cost NN O I-Metric
while NN O O
preserving NN O O
conversion NN O I-Metric
accuracy NN O I-Metric
. NN O O

We NN O O
conduct NN O O
experimental NN O O
evaluations NN O O
and NN O O
show NN O O
that NN O O
real-time NN O I-Method
VC NN O I-Method
is NN O O
capable NN O O
of NN O O
running NN O O
on NN O O
a NN O O
DSP NN O I-OtherScientificTerm
with NN O O
little NN O O
degradation NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
a NN O O
method NN O I-Generic
that NN O O
automatically NN O O
generates NN O O
paraphrase NN O I-OtherScientificTerm
sets NN O O
from NN O O
seed NN O O
sentences NN O O
to NN O O
be NN O O
used NN O O
as NN O O
reference NN O O
sets NN O O
in NN O O
objective NN O O
machine NN O I-Metric
translation NN O I-Metric
evaluation NN O I-Metric
measures NN O I-Metric
like NN O O
BLEU NN O I-Metric
and NN O O
NIST NN O I-Metric
. NN O O

We NN O O
measured NN O O
the NN O O
quality NN O O
of NN O O
the NN O O
paraphrases NN O I-OtherScientificTerm
produced NN O O
in NN O O
an NN O O
experiment NN O O
, NN O O
i.e. NN O O
, NN O O
-LRB- NN O O
i NN O O
-RRB- NN O O
their NN O O
grammaticality NN O I-Metric
: NN O O
at NN O O
least NN O O
99 NN O O
% NN O O
correct NN O O
sentences NN O O
; NN O O
-LRB- NN O O
ii NN O O
-RRB- NN O O
their NN O O
equivalence NN O I-Metric
in NN O I-Metric
meaning NN O I-Metric
: NN O O
at NN O O
least NN O O
96 NN O O
% NN O O
correct NN O O
paraphrases NN O I-OtherScientificTerm
either NN O O
by NN O O
meaning NN O I-Method
equivalence NN O I-Method
or NN O O
entailment NN O I-Method
; NN O O
and NN O O
, NN O O
-LRB- NN O O
iii NN O O
-RRB- NN O O
the NN O O
amount NN O O
of NN O O
internal NN O I-Metric
lexical NN O I-Metric
and NN O I-Metric
syntactical NN O I-Metric
variation NN O I-Metric
in NN O O
a NN O O
set NN O O
of NN O O
paraphrases NN O I-OtherScientificTerm
: NN O O
slightly NN O O
superior NN O O
to NN O O
that NN O O
of NN O O
hand-produced NN O I-OtherScientificTerm
sets NN O I-OtherScientificTerm
. NN O O

The NN O O
paraphrase NN O I-OtherScientificTerm
sets NN O O
produced NN O O
by NN O O
this NN O O
method NN O I-Generic
thus NN O O
seem NN O O
adequate NN O O
as NN O O
reference NN O O
sets NN O O
to NN O O
be NN O O
used NN O O
for NN O O
MT NN O I-Task
evaluation NN O I-Task
. NN O O

-DOCSTART- NN O O

Graph NN O I-Task
unification NN O I-Task
remains NN O O
the NN O O
most NN O O
expensive NN O O
part NN O O
of NN O O
unification-based NN O I-Task
grammar NN O I-Task
parsing NN O I-Task
. NN O O

We NN O O
focus NN O O
on NN O O
one NN O O
speed-up NN O I-Method
element NN O I-Method
in NN O O
the NN O O
design NN O O
of NN O O
unification NN O I-Method
algorithms NN O I-Method
: NN O O
avoidance NN O O
of NN O O
copying NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
unmodified NN O I-OtherScientificTerm
subgraphs NN O I-OtherScientificTerm
. NN O O

We NN O O
propose NN O O
a NN O O
method NN O I-Generic
of NN O O
attaining NN O O
such NN O O
a NN O O
design NN O O
through NN O O
a NN O O
method NN O O
of NN O O
structure-sharing NN O I-Method
which NN O O
avoids NN O O
log NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
d NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
overheads NN O I-OtherScientificTerm
often NN O O
associated NN O O
with NN O O
structure-sharing NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
graphs NN O I-OtherScientificTerm
without NN O O
any NN O O
use NN O O
of NN O O
costly NN O O
dependency NN O I-Method
pointers NN O I-Method
. NN O O

The NN O O
proposed NN O O
scheme NN O I-Generic
eliminates NN O O
redundant NN O I-OtherScientificTerm
copying NN O I-OtherScientificTerm
while NN O O
maintaining NN O O
the NN O O
quasi-destructive NN O I-OtherScientificTerm
scheme NN O I-OtherScientificTerm
's NN O I-OtherScientificTerm
ability NN O I-OtherScientificTerm
to NN O O
avoid NN O O
over NN O I-OtherScientificTerm
copying NN O I-OtherScientificTerm
and NN O O
early NN O I-OtherScientificTerm
copying NN O I-OtherScientificTerm
combined NN O O
with NN O O
its NN O O
ability NN O O
to NN O O
handle NN O O
cyclic NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
without NN O O
algorithmic NN O O
additions NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
a NN O O
novel NN O O
technique NN O O
and NN O O
implemented NN O O
system NN O I-Generic
for NN O O
constructing NN O O
a NN O O
subcategorization NN O I-OtherScientificTerm
dictionary NN O I-OtherScientificTerm
from NN O O
textual NN O I-Material
corpora NN O I-Material
. NN O O

Each NN O O
dictionary NN O O
entry NN O O
encodes NN O O
the NN O O
relative NN O I-OtherScientificTerm
frequency NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
occurrence NN O I-OtherScientificTerm
of NN O O
a NN O O
comprehensive NN O O
set NN O O
of NN O O
subcategorization NN O O
classes NN O O
for NN O O
English NN O O
. NN O O

An NN O O
initial NN O O
experiment NN O O
, NN O O
on NN O O
a NN O O
sample NN O O
of NN O O
14 NN O O
verbs NN O O
which NN O O
exhibit NN O O
multiple NN O O
complementation NN O O
patterns NN O O
, NN O O
demonstrates NN O O
that NN O O
the NN O O
technique NN O O
achieves NN O O
accuracy NN O O
comparable NN O O
to NN O O
previous NN O O
approaches NN O O
, NN O O
which NN O O
are NN O O
all NN O O
limited NN O O
to NN O O
a NN O O
highly NN O O
restricted NN O O
set NN O O
of NN O O
subcategorization NN O O
classes NN O O
. NN O O

We NN O O
also NN O O
demonstrate NN O O
that NN O O
a NN O O
subcategorization NN O I-OtherScientificTerm
dictionary NN O I-OtherScientificTerm
built NN O O
with NN O O
the NN O O
system NN O I-Generic
improves NN O O
the NN O O
accuracy NN O I-Metric
of NN O O
a NN O O
parser NN O I-Method
by NN O O
an NN O O
appreciable NN O O
amount NN O O

-DOCSTART- NN O O

The NN O O
introduction NN O O
of NN O O
prior NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
has NN O O
greatly NN O O
enhanced NN O O
numerous NN O O
purely NN O O
low-level NN O I-Method
driven NN O I-Method
image NN O I-Method
processing NN O I-Method
algorithms NN O I-Method
. NN O O

In NN O O
this NN O O
work NN O O
, NN O O
we NN O O
focus NN O O
on NN O O
the NN O O
problem NN O O
of NN O O
non-rigid NN O I-Task
image NN O I-Task
registration NN O I-Task
. NN O O

A NN O O
number NN O O
of NN O O
powerful NN O O
registration NN O I-Metric
criteria NN O I-Metric
have NN O O
been NN O O
developed NN O O
in NN O O
the NN O O
last NN O O
decade NN O O
, NN O O
most NN O O
prominently NN O O
the NN O O
criterion NN O O
of NN O O
maximum NN O I-Metric
mutual NN O I-Metric
information NN O I-Metric
. NN O O

Although NN O O
this NN O O
criterion NN O I-Generic
provides NN O O
for NN O O
good NN O O
registration NN O O
results NN O O
in NN O O
many NN O O
applications NN O O
, NN O O
it NN O I-Generic
remains NN O O
a NN O O
purely NN O O
low-level NN O I-Metric
criterion NN O I-Metric
. NN O O

As NN O O
a NN O O
consequence NN O O
, NN O O
registration NN O I-Task
results NN O O
will NN O O
deteriorate NN O O
once NN O O
this NN O O
low-level NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
is NN O O
corrupted NN O O
, NN O O
due NN O O
to NN O O
noise NN O I-OtherScientificTerm
, NN O O
partial NN O I-OtherScientificTerm
occlusions NN O I-OtherScientificTerm
or NN O O
missing NN O I-OtherScientificTerm
image NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
will NN O O
develop NN O O
a NN O O
Bayesian NN O I-Method
framework NN O I-Method
that NN O O
allows NN O O
to NN O O
impose NN O O
statistically NN O I-OtherScientificTerm
learned NN O I-OtherScientificTerm
prior NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
about NN O O
the NN O O
joint NN O I-OtherScientificTerm
intensity NN O I-OtherScientificTerm
distribution NN O I-OtherScientificTerm
into NN O O
image NN O I-Method
registration NN O I-Method
methods NN O I-Method
. NN O O

The NN O O
prior NN O I-Generic
is NN O O
given NN O O
by NN O O
a NN O O
kernel NN O I-Method
density NN O I-Method
estimate NN O I-Method
on NN O O
the NN O O
space NN O O
of NN O O
joint NN O I-OtherScientificTerm
intensity NN O I-OtherScientificTerm
distributions NN O I-OtherScientificTerm
computed NN O O
from NN O O
a NN O O
representative NN O O
set NN O O
of NN O O
pre-registered NN O I-Material
image NN O I-Material
pairs NN O I-Material
. NN O O

This NN O O
nonparametric NN O O
prior NN O O
accurately NN O O
models NN O O
previously NN O O
learned NN O O
intensity NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
between NN O O
various NN O O
image NN O I-OtherScientificTerm
modalities NN O I-OtherScientificTerm
and NN O O
slice NN O I-OtherScientificTerm
locations NN O I-OtherScientificTerm
. NN O O

Experimental NN O O
results NN O O
demonstrate NN O O
that NN O O
the NN O O
resulting NN O O
registration NN O I-Method
process NN O I-Method
is NN O O
more NN O O
robust NN O O
to NN O O
missing NN O I-OtherScientificTerm
low-level NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
as NN O O
it NN O I-OtherScientificTerm
favors NN O O
intensity NN O I-OtherScientificTerm
correspondences NN O I-OtherScientificTerm
statistically NN O O
consistent NN O O
with NN O O
the NN O O
learned NN O O
intensity NN O I-OtherScientificTerm
distributions NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

There NN O O
is NN O O
an NN O O
increased NN O O
interest NN O O
in NN O O
the NN O O
efficient NN O O
creation NN O I-Task
of NN O I-Task
city NN O I-Task
models NN O I-Task
, NN O O
be NN O O
it NN O O
virtual NN O O
or NN O O
as-built NN O O
. NN O O

We NN O O
present NN O O
a NN O O
method NN O I-Generic
for NN O O
synthesizing NN O I-Task
complex NN O I-Task
, NN O I-Task
photo-realistic NN O I-Task
facade NN O I-Task
images NN O I-Task
, NN O O
from NN O O
a NN O O
single NN O O
example NN O O
. NN O O

After NN O O
parsing NN O O
the NN O O
example NN O O
image NN O O
into NN O O
its NN O O
semantic NN O I-Method
components NN O I-Method
, NN O O
a NN O O
tiling NN O I-OtherScientificTerm
for NN O O
it NN O O
is NN O O
generated NN O O
. NN O O

Novel NN O O
tilings NN O I-OtherScientificTerm
can NN O O
then NN O O
be NN O O
created NN O O
, NN O O
yielding NN O O
facade NN O I-OtherScientificTerm
textures NN O I-OtherScientificTerm
with NN O O
different NN O O
dimensions NN O O
or NN O O
with NN O O
occluded NN O I-OtherScientificTerm
parts NN O I-OtherScientificTerm
inpainted NN O I-OtherScientificTerm
. NN O O

A NN O O
genetic NN O I-Generic
algorithm NN O I-Generic
guides NN O O
the NN O O
novel NN O O
facades NN O I-OtherScientificTerm
as NN O O
well NN O O
as NN O O
inpainted NN O I-OtherScientificTerm
parts NN O I-OtherScientificTerm
to NN O O
be NN O O
consistent NN O O
with NN O O
the NN O O
example NN O O
, NN O O
both NN O O
in NN O O
terms NN O O
of NN O O
their NN O O
overall NN O O
structure NN O O
and NN O O
their NN O O
detailed NN O O
textures NN O O
. NN O O

Promising NN O O
results NN O O
for NN O O
multiple NN O I-Generic
standard NN O I-Generic
datasets NN O I-Generic
-- NN O O
in NN O O
particular NN O O
for NN O O
the NN O O
different NN O O
building NN O O
styles NN O O
they NN O O
contain NN O O
-- NN O O
demonstrate NN O O
the NN O O
potential NN O O
of NN O O
the NN O O
method NN O I-Generic
. NN O O

-DOCSTART- NN O O

We NN O O
introduce NN O O
a NN O O
new NN O O
interactive NN O I-Method
corpus NN O I-Method
exploration NN O I-Method
tool NN O I-Method
called NN O O
InfoMagnets NN O I-Method
. NN O O

InfoMagnets NN O I-Method
aims NN O O
at NN O O
making NN O O
exploratory NN O I-Task
corpus NN O I-Task
analysis NN O I-Task
accessible NN O O
to NN O O
researchers NN O O
who NN O O
are NN O O
not NN O O
experts NN O O
in NN O O
text NN O I-Task
mining NN O I-Task
. NN O O

As NN O O
evidence NN O O
of NN O O
its NN O O
usefulness NN O O
and NN O O
usability NN O O
, NN O O
it NN O I-Generic
has NN O O
been NN O O
used NN O O
successfully NN O O
in NN O O
a NN O O
research NN O O
context NN O O
to NN O O
uncover NN O O
relationships NN O O
between NN O O
language NN O O
and NN O O
behavioral NN O O
patterns NN O O
in NN O O
two NN O O
distinct NN O O
domains NN O I-Generic
: NN O O
tutorial NN O I-Material
dialogue NN O I-Material
-LRB- NN O O
Kumar NN O O
et NN O O
al. NN O O
, NN O O
submitted NN O O
-RRB- NN O O
and NN O O
on-line NN O I-Material
communities NN O I-Material
-LRB- NN O O
Arguello NN O O
et NN O O
al. NN O O
, NN O O
2006 NN O O
-RRB- NN O O
. NN O O

As NN O O
an NN O O
educational NN O I-Method
tool NN O I-Method
, NN O O
it NN O I-Generic
has NN O O
been NN O O
used NN O O
as NN O O
part NN O O
of NN O O
a NN O O
unit NN O O
on NN O O
protocol NN O I-Task
analysis NN O I-Task
in NN O O
an NN O O
Educational NN O O
Research NN O O
Methods NN O O
course NN O O
. NN O O

-DOCSTART- NN O O

Sources NN O O
of NN O O
training NN O O
data NN O O
suitable NN O O
for NN O O
language NN O I-Method
modeling NN O I-Method
of NN O O
conversational NN O I-Material
speech NN O I-Material
are NN O O
limited NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
show NN O O
how NN O O
training NN O O
data NN O O
can NN O O
be NN O O
supplemented NN O O
with NN O O
text NN O O
from NN O O
the NN O O
web NN O O
filtered NN O O
to NN O O
match NN O O
the NN O O
style NN O O
and/or NN O O
topic NN O O
of NN O O
the NN O O
target NN O O
recognition NN O I-Task
task NN O I-Task
, NN O O
but NN O O
also NN O O
that NN O O
it NN O O
is NN O O
possible NN O O
to NN O O
get NN O O
bigger NN O O
performance NN O O
gains NN O O
from NN O O
the NN O O
data NN O O
by NN O O
using NN O O
class-dependent NN O I-Method
interpolation NN O I-Method
of NN O I-Method
N-grams NN O I-Method
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
a NN O O
method NN O I-Generic
for NN O O
detecting NN O I-Task
3D NN O I-Task
objects NN O I-Task
using NN O O
multi-modalities NN O I-OtherScientificTerm
. NN O O

While NN O O
it NN O I-Generic
is NN O O
generic NN O O
, NN O O
we NN O O
demonstrate NN O O
it NN O I-Generic
on NN O O
the NN O O
combination NN O O
of NN O O
an NN O O
image NN O I-OtherScientificTerm
and NN O O
a NN O O
dense NN O I-OtherScientificTerm
depth NN O I-OtherScientificTerm
map NN O I-OtherScientificTerm
which NN O O
give NN O O
complementary NN O I-OtherScientificTerm
object NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
. NN O O

It NN O I-Generic
works NN O O
in NN O O
real-time NN O O
, NN O O
under NN O O
heavy NN O O
clutter NN O O
, NN O O
does NN O O
not NN O O
require NN O O
a NN O O
time NN O I-OtherScientificTerm
consuming NN O I-OtherScientificTerm
training NN O I-OtherScientificTerm
stage NN O I-OtherScientificTerm
, NN O O
and NN O O
can NN O O
handle NN O O
untextured NN O I-OtherScientificTerm
objects NN O I-OtherScientificTerm
. NN O O

It NN O I-Generic
is NN O O
based NN O O
on NN O O
an NN O O
efficient NN O O
representation NN O O
of NN O O
templates NN O I-OtherScientificTerm
that NN O O
capture NN O O
the NN O O
different NN O O
modalities NN O I-OtherScientificTerm
, NN O O
and NN O O
we NN O O
show NN O O
in NN O O
many NN O O
experiments NN O O
on NN O O
commodity NN O I-OtherScientificTerm
hardware NN O I-OtherScientificTerm
that NN O O
our NN O O
approach NN O I-Generic
significantly NN O O
outperforms NN O O
state-of-the-art NN O I-Generic
methods NN O I-Generic
on NN O O
single NN O I-OtherScientificTerm
modalities NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

The NN O O
compact NN O I-Task
description NN O I-Task
of NN O I-Task
a NN O I-Task
video NN O I-Task
sequence NN O I-Task
through NN O O
a NN O O
single NN O O
image NN O I-OtherScientificTerm
map NN O I-OtherScientificTerm
and NN O O
a NN O O
dominant NN O I-OtherScientificTerm
motion NN O I-OtherScientificTerm
has NN O O
applications NN O O
in NN O O
several NN O O
domains NN O I-Generic
, NN O O
including NN O O
video NN O I-Task
browsing NN O I-Task
and NN O I-Task
retrieval NN O I-Task
, NN O O
compression NN O I-Task
, NN O O
mosaicing NN O I-Task
, NN O O
and NN O O
visual NN O I-Task
summarization NN O I-Task
. NN O O

Building NN O O
such NN O O
a NN O O
representation NN O I-Generic
requires NN O O
the NN O O
capability NN O O
to NN O O
register NN O O
all NN O O
the NN O O
frames NN O O
with NN O O
respect NN O O
to NN O O
the NN O O
dominant NN O O
object NN O O
in NN O O
the NN O O
scene NN O O
, NN O O
a NN O O
task NN O I-Generic
which NN O O
has NN O O
been NN O O
, NN O O
in NN O O
the NN O O
past NN O O
, NN O O
addressed NN O O
through NN O O
temporally NN O O
localized NN O I-Method
motion NN O I-Method
estimates NN O I-Method
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
show NN O O
how NN O O
the NN O O
lack NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
temporal NN O I-OtherScientificTerm
consistency NN O I-OtherScientificTerm
associated NN O O
with NN O O
such NN O O
estimates NN O I-Generic
can NN O O
undermine NN O O
the NN O O
validity NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
dominant NN O I-OtherScientificTerm
motion NN O I-OtherScientificTerm
assumption NN O I-OtherScientificTerm
, NN O O
leading NN O O
to NN O O
oscillation NN O I-OtherScientificTerm
between NN O I-OtherScientificTerm
different NN O I-OtherScientificTerm
scene NN O I-OtherScientificTerm
interpretations NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
poor NN O I-OtherScientificTerm
registration NN O I-OtherScientificTerm
. NN O O

To NN O O
avoid NN O O
this NN O O
oscillation NN O I-Generic
, NN O O
we NN O O
augment NN O O
the NN O O
motion NN O I-Method
model NN O I-Method
with NN O O
a NN O O
generic NN O I-OtherScientificTerm
temporal NN O I-OtherScientificTerm
constraint NN O I-OtherScientificTerm
which NN O O
increases NN O O
the NN O O
robustness NN O I-Metric
against NN O O
competing NN O O
interpretations NN O O
, NN O O
leading NN O O
to NN O O
more NN O O
meaningful NN O O
content NN O I-Task
summarization NN O I-Task
. NN O O

-DOCSTART- NN O O

In NN O O
cross-domain NN O I-Task
learning NN O I-Task
, NN O O
there NN O O
is NN O O
a NN O O
more NN O O
challenging NN O O
problem NN O O
that NN O O
the NN O O
domain NN O I-OtherScientificTerm
divergence NN O I-OtherScientificTerm
involves NN O O
more NN O O
than NN O O
one NN O O
dominant NN O I-Generic
factors NN O I-Generic
, NN O O
e.g. NN O O
, NN O O
different NN O O
viewpoints NN O I-OtherScientificTerm
, NN O O
various NN O O
resolutions NN O I-OtherScientificTerm
and NN O O
changing NN O O
illuminations NN O I-OtherScientificTerm
. NN O O

Fortunately NN O O
, NN O O
an NN O O
intermediate NN O I-OtherScientificTerm
domain NN O I-OtherScientificTerm
could NN O O
often NN O O
be NN O O
found NN O O
to NN O O
build NN O O
a NN O O
bridge NN O O
across NN O O
them NN O O
to NN O O
facilitate NN O O
the NN O O
learning NN O I-Task
problem NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
Coupled NN O I-Method
Marginalized NN O I-Method
Denoising NN O I-Method
Auto-encoders NN O I-Method
framework NN O I-Method
to NN O O
address NN O O
the NN O O
cross-domain NN O I-Task
problem NN O I-Task
. NN O O

Specifically NN O O
, NN O O
we NN O O
design NN O O
two NN O O
marginalized NN O I-Method
denoising NN O I-Method
auto-encoders NN O I-Method
, NN O O
one NN O I-Generic
for NN O O
the NN O O
target NN O O
and NN O O
the NN O O
other NN O I-Generic
for NN O O
source NN O O
as NN O O
well NN O O
as NN O O
the NN O O
intermediate NN O O
one NN O O
. NN O O

To NN O O
better NN O O
couple NN O O
the NN O O
two NN O O
denoising NN O I-Method
auto-encoders NN O I-Method
learning NN O I-Method
, NN O O
we NN O O
incorporate NN O O
a NN O O
feature NN O I-Method
mapping NN O I-Method
, NN O O
which NN O O
tends NN O O
to NN O O
transfer NN O O
knowledge NN O O
between NN O O
the NN O O
intermediate NN O I-OtherScientificTerm
domain NN O I-OtherScientificTerm
and NN O O
the NN O O
target NN O O
one NN O O
. NN O O

Furthermore NN O O
, NN O O
the NN O O
maximum NN O I-OtherScientificTerm
margin NN O I-OtherScientificTerm
criterion NN O I-OtherScientificTerm
, NN O O
e.g. NN O O
, NN O O
intra-class NN O I-OtherScientificTerm
com-pactness NN O I-OtherScientificTerm
and NN O O
inter-class NN O I-OtherScientificTerm
penalty NN O I-OtherScientificTerm
, NN O O
on NN O O
the NN O O
output NN O O
layer NN O O
is NN O O
imposed NN O O
to NN O O
seek NN O O
more NN O O
discriminative NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
across NN O O
different NN O O
domains NN O O
. NN O O

Extensive NN O O
experiments NN O O
on NN O O
two NN O O
tasks NN O I-Generic
have NN O O
demonstrated NN O O
the NN O O
superiority NN O O
of NN O O
our NN O O
method NN O I-Generic
over NN O O
the NN O O
state-of-the-art NN O I-Generic
methods NN O I-Generic
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
aim NN O O
to NN O O
automatically NN O O
render NN O O
aging NN O I-OtherScientificTerm
faces NN O I-OtherScientificTerm
in NN O O
a NN O O
personalized NN O O
way NN O O
. NN O O

Basically NN O O
, NN O O
a NN O O
set NN O O
of NN O O
age-group NN O I-Method
specific NN O I-Method
dictionaries NN O I-Method
are NN O O
learned NN O O
, NN O O
where NN O O
the NN O O
dictionary NN O I-OtherScientificTerm
bases NN O I-OtherScientificTerm
corresponding NN O O
to NN O O
the NN O O
same NN O O
index NN O O
yet NN O O
from NN O O
different NN O O
dictionaries NN O O
form NN O O
a NN O O
particular NN O O
aging NN O I-OtherScientificTerm
process NN O I-OtherScientificTerm
pattern NN O I-OtherScientificTerm
cross NN O O
different NN O O
age NN O O
groups NN O O
, NN O O
and NN O O
a NN O O
linear NN O I-Method
combination NN O I-Method
of NN O O
these NN O O
patterns NN O I-Generic
expresses NN O O
a NN O O
particular NN O O
personalized NN O I-Task
aging NN O I-Task
process NN O I-Task
. NN O O

Moreover NN O O
, NN O O
two NN O O
factors NN O O
are NN O O
taken NN O O
into NN O O
consideration NN O O
in NN O O
the NN O O
dictionary NN O I-Method
learning NN O I-Method
process NN O I-Method
. NN O O

First NN O O
, NN O O
beyond NN O O
the NN O O
aging NN O I-Method
dictionaries NN O I-Method
, NN O O
each NN O O
subject NN O O
may NN O O
have NN O O
extra NN O O
personalized NN O I-OtherScientificTerm
facial NN O I-OtherScientificTerm
characteristics NN O I-OtherScientificTerm
, NN O O
e.g. NN O O
mole NN O I-OtherScientificTerm
, NN O O
which NN O O
are NN O O
invariant NN O O
in NN O O
the NN O O
aging NN O I-Task
process NN O I-Task
. NN O O

Second NN O O
, NN O O
it NN O O
is NN O O
challenging NN O O
or NN O O
even NN O O
impossible NN O O
to NN O O
collect NN O O
faces NN O O
of NN O O
all NN O O
age NN O O
groups NN O O
for NN O O
a NN O O
particular NN O O
subject NN O O
, NN O O
yet NN O O
much NN O O
easier NN O O
and NN O O
more NN O O
practical NN O O
to NN O O
get NN O O
face NN O O
pairs NN O O
from NN O O
neighboring NN O O
age NN O O
groups NN O O
. NN O O

Thus NN O O
a NN O O
personality-aware NN O I-OtherScientificTerm
coupled NN O I-OtherScientificTerm
reconstruction NN O I-OtherScientificTerm
loss NN O I-OtherScientificTerm
is NN O O
utilized NN O O
to NN O O
learn NN O O
the NN O O
dictionaries NN O I-Generic
based NN O O
on NN O O
face NN O O
pairs NN O O
from NN O O
neighboring NN O O
age NN O O
groups NN O O
. NN O O

Extensive NN O O
experiments NN O O
well NN O O
demonstrate NN O O
the NN O O
advantages NN O O
of NN O O
our NN O O
proposed NN O O
solution NN O I-Generic
over NN O O
other NN O O
state-of-the-arts NN O I-Generic
in NN O O
term NN O O
of NN O O
personalized NN O I-OtherScientificTerm
aging NN O I-OtherScientificTerm
progression NN O I-OtherScientificTerm
, NN O O
as NN O O
well NN O O
as NN O O
the NN O O
performance NN O O
gain NN O O
for NN O O
cross-age NN O I-Task
face NN O I-Task
verification NN O I-Task
by NN O O
synthesizing NN O I-Method
aging NN O I-Method
faces NN O I-Method
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
a NN O O
draft NN O O
scheme NN O O
of NN O O
the NN O O
model NN O I-Generic
formalizing NN O O
the NN O O
structure NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
communicative NN O I-OtherScientificTerm
context NN O I-OtherScientificTerm
in NN O O
dialogue NN O I-Material
interaction NN O I-Material
. NN O O

The NN O O
relationships NN O O
between NN O O
the NN O O
interacting NN O O
partners NN O O
are NN O O
considered NN O O
as NN O O
system NN O O
of NN O O
three NN O O
automata NN O O
representing NN O O
the NN O O
partners NN O O
of NN O O
the NN O O
dialogue NN O O
and NN O O
environment NN O O
. NN O O

-DOCSTART- NN O O

Visitors NN O O
who NN O O
browse NN O O
the NN O O
web NN O O
from NN O O
wireless NN O O
PDAs NN O O
, NN O O
cell NN O O
phones NN O O
, NN O O
and NN O O
pagers NN O O
are NN O O
frequently NN O O
stymied NN O O
by NN O O
web NN O I-Method
interfaces NN O I-Method
optimized NN O O
for NN O O
desktop NN O I-OtherScientificTerm
PCs NN O I-OtherScientificTerm
. NN O O

Simply NN O O
replacing NN O O
graphics NN O O
with NN O O
text NN O O
and NN O O
reformatting NN O O
tables NN O O
does NN O O
not NN O O
solve NN O O
the NN O O
problem NN O O
, NN O O
because NN O O
deep NN O I-OtherScientificTerm
link NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
can NN O O
still NN O O
require NN O O
minutes NN O O
to NN O O
traverse NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
develop NN O O
an NN O O
algorithm NN O I-Generic
, NN O O
MINPATH NN O I-Method
, NN O O
that NN O O
automatically NN O O
improves NN O O
wireless NN O I-Task
web NN O I-Task
navigation NN O I-Task
by NN O O
suggesting NN O O
useful NN O O
shortcut NN O O
links NN O O
in NN O O
real NN O O
time NN O O
. NN O O

MINPATH NN O I-Method
finds NN O O
shortcuts NN O O
by NN O O
using NN O O
a NN O O
learned NN O O
model NN O I-Generic
of NN O O
web NN O I-OtherScientificTerm
visitor NN O I-OtherScientificTerm
behavior NN O I-OtherScientificTerm
to NN O O
estimate NN O O
the NN O O
savings NN O I-Task
of NN O I-Task
shortcut NN O I-Task
links NN O I-Task
, NN O O
and NN O O
suggests NN O O
only NN O O
the NN O O
few NN O O
best NN O O
links NN O O
. NN O O

We NN O O
explore NN O O
a NN O O
variety NN O O
of NN O O
predictive NN O I-Generic
models NN O I-Generic
, NN O O
including NN O O
Na NN O I-Method
¨ NN O I-Method
ıve NN O I-Method
Bayes NN O I-Method
mixture NN O I-Method
models NN O I-Method
and NN O O
mixtures NN O I-Method
of NN O I-Method
Markov NN O I-Method
models NN O I-Method
, NN O O
and NN O O
report NN O O
empirical NN O O
evidence NN O O
that NN O O
MINPATH NN O I-Method
finds NN O O
useful NN O O
shortcuts NN O O
that NN O O
save NN O O
substantial NN O O
navigational NN O O
effort NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
a NN O O
particular NN O O
approach NN O I-Generic
to NN O O
parsing NN O I-Task
that NN O O
utilizes NN O O
recent NN O O
advances NN O O
in NN O O
unification-based NN O I-Task
parsing NN O I-Task
and NN O O
in NN O O
classification-based NN O I-Task
knowledge NN O I-Task
representation NN O I-Task
. NN O O

As NN O O
unification-based NN O I-Method
grammatical NN O I-Method
frameworks NN O I-Method
are NN O O
extended NN O O
to NN O O
handle NN O O
richer NN O O
descriptions NN O O
of NN O O
linguistic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
, NN O O
they NN O I-Generic
begin NN O O
to NN O O
share NN O O
many NN O O
of NN O O
the NN O O
properties NN O O
that NN O O
have NN O O
been NN O O
developed NN O O
in NN O O
KL-ONE-like NN O I-Method
knowledge NN O I-Method
representation NN O I-Method
systems NN O I-Method
. NN O O

This NN O O
commonality NN O O
suggests NN O O
that NN O O
some NN O O
of NN O O
the NN O O
classification-based NN O I-Method
representation NN O I-Method
techniques NN O I-Method
can NN O O
be NN O O
applied NN O O
to NN O O
unification-based NN O I-OtherScientificTerm
linguistic NN O I-OtherScientificTerm
descriptions NN O I-OtherScientificTerm
. NN O O

This NN O O
merging NN O O
supports NN O O
the NN O O
integration NN O O
of NN O O
semantic NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
syntactic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
into NN O O
the NN O O
same NN O O
system NN O I-Generic
, NN O O
simultaneously NN O O
subject NN O O
to NN O O
the NN O O
same NN O O
types NN O O
of NN O O
processes NN O O
, NN O O
in NN O O
an NN O O
efficient NN O O
manner NN O O
. NN O O

The NN O O
result NN O O
is NN O O
expected NN O O
to NN O O
be NN O O
more NN O O
efficient NN O O
parsing NN O I-Method
due NN O O
to NN O O
the NN O O
increased NN O O
organization NN O O
of NN O O
knowledge NN O O
. NN O O

The NN O O
use NN O O
of NN O O
a NN O O
KL-ONE NN O I-Method
style NN O I-Method
representation NN O I-Method
for NN O O
parsing NN O I-Task
and NN O O
semantic NN O I-Task
interpretation NN O I-Task
was NN O O
first NN O O
explored NN O O
in NN O O
the NN O O
PSI-KLONE NN O I-Method
system NN O I-Method
-LSB- NN O O
2 NN O O
-RSB- NN O O
, NN O O
in NN O O
which NN O O
parsing NN O I-Task
is NN O O
characterized NN O O
as NN O O
an NN O O
inference NN O I-Method
process NN O I-Method
called NN O O
incremental NN O I-Method
description NN O I-Method
refinement NN O I-Method
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
discuss NN O O
a NN O O
proposed NN O O
user NN O I-Method
knowledge NN O I-Method
modeling NN O I-Method
architecture NN O I-Method
for NN O O
the NN O O
ICICLE NN O I-Task
system NN O I-Task
, NN O O
a NN O O
language NN O I-Task
tutoring NN O I-Task
application NN O I-Task
for NN O O
deaf NN O I-OtherScientificTerm
learners NN O I-OtherScientificTerm
of NN O O
written NN O I-Material
English NN O I-Material
. NN O O

The NN O O
model NN O I-Generic
will NN O O
represent NN O O
the NN O O
language NN O O
proficiency NN O O
of NN O O
the NN O O
user NN O O
and NN O O
is NN O O
designed NN O O
to NN O O
be NN O O
referenced NN O O
during NN O O
both NN O O
writing NN O I-Task
analysis NN O I-Task
and NN O O
feedback NN O I-Task
production NN O I-Task
. NN O O

We NN O O
motivate NN O O
our NN O O
model NN O I-Generic
design NN O I-Generic
by NN O O
citing NN O O
relevant NN O O
research NN O O
on NN O O
second NN O I-Task
language NN O I-Task
and NN O I-Task
cognitive NN O I-Task
skill NN O I-Task
acquisition NN O I-Task
, NN O O
and NN O O
briefly NN O O
discuss NN O O
preliminary NN O O
empirical NN O O
evidence NN O O
supporting NN O O
the NN O O
design NN O I-Generic
. NN O O

We NN O O
conclude NN O O
by NN O O
showing NN O O
how NN O O
our NN O O
design NN O I-Generic
can NN O O
provide NN O O
a NN O O
rich NN O O
and NN O O
robust NN O O
information NN O O
base NN O O
to NN O O
a NN O O
language NN O I-Task
assessment NN O I-Task
/ NN O I-Task
correction NN O I-Task
application NN O I-Task
by NN O O
modeling NN O O
user NN O I-OtherScientificTerm
proficiency NN O I-OtherScientificTerm
at NN O O
a NN O O
high NN O O
level NN O O
of NN O O
granularity NN O I-Metric
and NN O O
specificity NN O I-Metric
. NN O O

-DOCSTART- NN O O

Constraint NN O I-Method
propagation NN O I-Method
is NN O O
one NN O O
of NN O O
the NN O O
key NN O O
techniques NN O O
in NN O O
constraint NN O I-Method
programming NN O I-Method
, NN O O
and NN O O
a NN O O
large NN O O
body NN O O
of NN O O
work NN O O
has NN O O
built NN O O
up NN O O
around NN O O
it NN O O
. NN O O

Special-purpose NN O I-Method
constraint NN O I-Method
propagation NN O I-Method
algorithms NN O I-Method
frequently NN O O
make NN O O
implicit NN O O
use NN O O
of NN O O
short NN O O
supports NN O O
-- NN O O
by NN O O
examining NN O O
a NN O O
subset NN O O
of NN O O
the NN O O
variables NN O O
, NN O O
they NN O O
can NN O O
infer NN O O
support NN O O
-LRB- NN O O
a NN O O
justification NN O O
that NN O O
a NN O O
variable-value NN O O
pair NN O O
still NN O O
forms NN O O
part NN O O
of NN O O
a NN O O
solution NN O O
to NN O O
the NN O O
constraint NN O O
-RRB- NN O O
for NN O O
all NN O O
other NN O O
variables NN O O
and NN O O
values NN O O
and NN O O
save NN O O
substantial NN O O
work NN O O
. NN O O

Recently NN O O
short NN O O
supports NN O O
have NN O O
been NN O O
used NN O O
in NN O O
general NN O I-Method
purpose NN O I-Method
prop-agators NN O I-Method
, NN O O
and NN O O
-LRB- NN O O
when NN O O
the NN O O
constraint NN O I-OtherScientificTerm
is NN O O
amenable NN O O
to NN O O
short NN O O
supports NN O O
-RRB- NN O O
speed NN O O
ups NN O O
of NN O O
more NN O O
than NN O O
three NN O O
orders NN O O
of NN O O
magnitude NN O O
have NN O O
been NN O O
demonstrated NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
present NN O O
SHORTSTR2 NN O I-Method
, NN O O
a NN O O
development NN O O
of NN O O
the NN O O
Simple NN O I-Method
Tabular NN O I-Method
Reduction NN O I-Method
algorithm NN O I-Method
STR2 NN O I-Method
+ NN O I-Method
. NN O O

We NN O O
show NN O O
that NN O O
SHORTSTR2 NN O I-Method
is NN O O
complementary NN O O
to NN O O
the NN O O
existing NN O O
algorithms NN O O
SHORTGAC NN O I-Method
and NN O O
HAGGISGAC NN O I-Method
that NN O O
exploit NN O O
short NN O O
supports NN O O
, NN O O
while NN O O
being NN O O
much NN O O
simpler NN O O
. NN O O

When NN O O
a NN O O
constraint NN O I-OtherScientificTerm
is NN O O
amenable NN O O
to NN O O
short NN O I-OtherScientificTerm
supports NN O I-OtherScientificTerm
, NN O O
the NN O O
short NN O I-OtherScientificTerm
support NN O I-OtherScientificTerm
set NN O I-OtherScientificTerm
can NN O O
be NN O O
exponentially NN O O
smaller NN O O
than NN O O
the NN O O
full-length NN O I-OtherScientificTerm
support NN O I-OtherScientificTerm
set NN O I-OtherScientificTerm
. NN O O

Therefore NN O O
SHORTSTR2 NN O I-Method
can NN O O
efficiently NN O O
propagate NN O O
many NN O O
constraints NN O I-OtherScientificTerm
that NN O O
STR2 NN O I-Method
+ NN O I-Method
can NN O O
not NN O O
even NN O O
load NN O O
into NN O O
memory NN O O
. NN O O

We NN O O
also NN O O
show NN O O
that NN O O
SHORTSTR2 NN O I-Method
can NN O O
be NN O O
combined NN O O
with NN O O
a NN O O
simple NN O O
algorithm NN O I-Generic
to NN O O
identify NN O O
short NN O I-OtherScientificTerm
supports NN O I-OtherScientificTerm
from NN O O
full-length NN O I-OtherScientificTerm
supports NN O I-OtherScientificTerm
, NN O O
to NN O O
provide NN O O
a NN O O
superior NN O O
drop-in NN O I-OtherScientificTerm
replacement NN O I-OtherScientificTerm
for NN O O
STR2 NN O I-Method
+ NN O I-Method
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
a NN O O
detection NN O I-Method
method NN O I-Method
for NN O O
orthographic NN O I-OtherScientificTerm
variants NN O I-OtherScientificTerm
caused NN O O
by NN O O
transliteration NN O I-Task
in NN O O
a NN O O
large NN O O
corpus NN O O
. NN O O

The NN O O
method NN O I-Generic
employs NN O O
two NN O O
similarities NN O I-Generic
. NN O O

One NN O O
is NN O O
string NN O I-OtherScientificTerm
similarity NN O I-OtherScientificTerm
based NN O O
on NN O O
edit NN O I-Method
distance NN O I-Method
. NN O O

The NN O O
other NN O O
is NN O O
contextual NN O I-OtherScientificTerm
similarity NN O I-OtherScientificTerm
by NN O O
a NN O O
vector NN O I-Method
space NN O I-Method
model NN O I-Method
. NN O O

Experimental NN O O
results NN O O
show NN O O
that NN O O
the NN O O
method NN O I-Generic
performed NN O O
a NN O O
0.889 NN O O
F-measure NN O I-Metric
in NN O O
an NN O O
open NN O O
test NN O O
. NN O O

-DOCSTART- NN O O

Uncertainty NN O I-Task
handling NN O I-Task
plays NN O O
an NN O O
important NN O O
role NN O O
during NN O O
shape NN O I-Task
tracking NN O I-Task
. NN O O

We NN O O
have NN O O
recently NN O O
shown NN O O
that NN O O
the NN O O
fusion NN O I-Method
of NN O I-Method
measurement NN O I-Method
information NN O I-Method
with NN O I-Method
system NN O I-Method
dynamics NN O I-Method
and NN O I-Method
shape NN O I-Method
priors NN O I-Method
greatly NN O O
improves NN O O
the NN O O
tracking NN O I-Task
performance NN O O
for NN O O
very NN O O
noisy NN O I-Material
images NN O I-Material
such NN O O
as NN O O
ultrasound NN O I-Material
sequences NN O I-Material
-LSB- NN O O
22 NN O O
-RSB- NN O O
. NN O O

Nevertheless NN O O
, NN O O
this NN O O
approach NN O I-Generic
required NN O O
user NN O I-OtherScientificTerm
initialization NN O I-OtherScientificTerm
of NN O O
the NN O O
tracking NN O I-Method
process NN O I-Method
. NN O O

This NN O O
paper NN O O
solves NN O O
the NN O O
automatic NN O I-Task
initial-ization NN O I-Task
problem NN O I-Task
by NN O O
performing NN O O
boosted NN O I-Method
shape NN O I-Method
detection NN O I-Method
as NN O O
a NN O O
generic NN O I-Method
measurement NN O I-Method
process NN O I-Method
and NN O O
integrating NN O O
it NN O I-Generic
in NN O O
our NN O O
tracking NN O I-Method
framework NN O I-Method
. NN O O

We NN O O
show NN O O
how NN O O
to NN O O
propagate NN O O
the NN O O
local NN O I-OtherScientificTerm
detection NN O I-OtherScientificTerm
uncertainties NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
multiple NN O I-OtherScientificTerm
shape NN O I-OtherScientificTerm
candidates NN O I-OtherScientificTerm
during NN O O
shape NN O I-OtherScientificTerm
alignment NN O I-OtherScientificTerm
, NN O O
fusion NN O O
with NN O O
the NN O O
predicted NN O I-OtherScientificTerm
shape NN O I-OtherScientificTerm
prior NN O I-OtherScientificTerm
, NN O O
and NN O O
fusion NN O O
with NN O O
subspace NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
. NN O O

As NN O O
a NN O O
result NN O O
, NN O O
we NN O O
treat NN O O
all NN O O
sources NN O O
of NN O O
information NN O O
in NN O O
a NN O O
unified NN O O
way NN O O
and NN O O
derive NN O O
the NN O O
posterior NN O I-Method
shape NN O I-Method
model NN O I-Method
as NN O O
the NN O O
shape NN O O
with NN O O
the NN O O
maximum NN O I-OtherScientificTerm
likelihood NN O I-OtherScientificTerm
. NN O O

Our NN O O
framework NN O I-Generic
is NN O O
applied NN O O
for NN O O
the NN O O
automatic NN O I-Task
tracking NN O I-Task
of NN O I-Task
endocardium NN O I-OtherScientificTerm
in NN O O
ultrasound NN O I-Material
sequences NN O I-Material
of NN O I-Material
the NN O I-Material
human NN O I-Material
heart NN O I-Material
. NN O O

Reliable NN O O
detection NN O I-Task
and NN O O
robust NN O O
tracking NN O I-Task
results NN O O
are NN O O
achieved NN O O
when NN O O
compared NN O O
to NN O O
existing NN O O
approaches NN O I-Generic
and NN O O
inter-expert NN O I-OtherScientificTerm
variations NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
a NN O O
syntax-based NN O I-OtherScientificTerm
constraint NN O I-OtherScientificTerm
for NN O O
word NN O I-Task
alignment NN O I-Task
, NN O O
known NN O O
as NN O O
the NN O O
cohesion NN O I-OtherScientificTerm
constraint NN O I-OtherScientificTerm
. NN O O

It NN O I-Generic
requires NN O O
disjoint NN O O
English NN O I-Material
phrases NN O I-Material
to NN O O
be NN O O
mapped NN O O
to NN O O
non-overlapping NN O O
intervals NN O O
in NN O O
the NN O O
French NN O I-Material
sentence NN O I-Material
. NN O O

We NN O O
evaluate NN O O
the NN O O
utility NN O O
of NN O O
this NN O O
constraint NN O I-Generic
in NN O O
two NN O O
different NN O O
algorithms NN O I-Generic
. NN O O

The NN O O
results NN O O
show NN O O
that NN O O
it NN O I-Generic
can NN O O
provide NN O O
a NN O O
significant NN O O
improvement NN O O
in NN O O
alignment NN O I-Metric
quality NN O I-Metric
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
considers NN O O
the NN O O
problem NN O O
of NN O O
automatic NN O I-Task
assessment NN O I-Task
of NN O I-Task
local NN O I-Task
coherence NN O I-Task
. NN O O

We NN O O
present NN O O
a NN O O
novel NN O O
entity-based NN O I-Method
representation NN O I-Method
of NN O I-Method
discourse NN O I-Method
which NN O O
is NN O O
inspired NN O O
by NN O O
Centering NN O I-Method
Theory NN O I-Method
and NN O O
can NN O O
be NN O O
computed NN O O
automatically NN O O
from NN O O
raw NN O I-Material
text NN O I-Material
. NN O O

We NN O O
view NN O O
coherence NN O I-Task
assessment NN O I-Task
as NN O O
a NN O O
ranking NN O I-Task
learning NN O I-Task
problem NN O I-Task
and NN O O
show NN O O
that NN O O
the NN O O
proposed NN O O
discourse NN O I-Method
representation NN O I-Method
supports NN O O
the NN O O
effective NN O O
learning NN O O
of NN O O
a NN O O
ranking NN O I-OtherScientificTerm
function NN O I-OtherScientificTerm
. NN O O

Our NN O O
experiments NN O O
demonstrate NN O O
that NN O O
the NN O O
induced NN O I-Method
model NN O I-Method
achieves NN O O
significantly NN O O
higher NN O O
accuracy NN O I-Metric
than NN O O
a NN O O
state-of-the-art NN O O
coherence NN O I-Method
model NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
introduces NN O O
a NN O O
robust NN O I-Generic
interactive NN O I-Generic
method NN O I-Generic
for NN O O
speech NN O I-Task
understanding NN O I-Task
. NN O O

The NN O O
generalized NN O I-Method
LR NN O I-Method
parsing NN O I-Method
is NN O O
enhanced NN O O
in NN O O
this NN O O
approach NN O I-Generic
. NN O O

Parsing NN O I-Method
proceeds NN O O
from NN O O
left NN O O
to NN O O
right NN O O
correcting NN O O
minor NN O O
errors NN O O
. NN O O

When NN O O
a NN O O
very NN O O
noisy NN O O
portion NN O O
is NN O O
detected NN O O
, NN O O
the NN O O
parser NN O I-Method
skips NN O O
that NN O O
portion NN O O
using NN O O
a NN O O
fake NN O O
non-terminal NN O I-OtherScientificTerm
symbol NN O I-OtherScientificTerm
. NN O O

The NN O O
unidentified NN O O
portion NN O O
is NN O O
resolved NN O O
by NN O O
re-utterance NN O O
of NN O O
that NN O O
portion NN O O
which NN O O
is NN O O
parsed NN O O
very NN O O
efficiently NN O O
by NN O O
using NN O O
the NN O O
parse NN O O
record NN O O
of NN O O
the NN O O
first NN O O
utterance NN O O
. NN O O

The NN O O
user NN O O
does NN O O
not NN O O
have NN O O
to NN O O
speak NN O O
the NN O O
whole NN O O
sentence NN O O
again NN O O
. NN O O

This NN O O
method NN O I-Generic
is NN O O
also NN O O
capable NN O O
of NN O O
handling NN O O
unknown NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
, NN O O
which NN O O
is NN O O
important NN O O
in NN O O
practical NN O O
systems NN O O
. NN O O

Detected NN O O
unknown NN O O
words NN O O
can NN O O
be NN O O
incrementally NN O O
incorporated NN O O
into NN O O
the NN O O
dictionary NN O O
after NN O O
the NN O O
interaction NN O O
with NN O O
the NN O O
user NN O O
. NN O O

A NN O O
pilot NN O O
system NN O O
has NN O O
shown NN O O
great NN O O
effectiveness NN O O
of NN O O
this NN O O
approach NN O I-Generic
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
shows NN O O
that NN O O
it NN O O
is NN O O
very NN O O
often NN O O
possible NN O O
to NN O O
identify NN O O
the NN O O
source NN O O
language NN O O
of NN O O
medium-length NN O I-Material
speeches NN O I-Material
in NN O O
the NN O O
EUROPARL NN O I-Material
corpus NN O I-Material
on NN O O
the NN O O
basis NN O O
of NN O O
frequency NN O I-Method
counts NN O I-Method
of NN O I-Method
word NN O I-OtherScientificTerm
n-grams NN O I-OtherScientificTerm
-LRB- NN O O
87.2 NN O O
% NN O O
-96.7 NN O O
% NN O O
accuracy NN O I-Metric
depending NN O O
on NN O O
classification NN O I-Method
method NN O I-Method
-RRB- NN O O
. NN O O

The NN O O
paper NN O O
also NN O O
examines NN O O
in NN O O
detail NN O O
which NN O O
positive NN O I-OtherScientificTerm
markers NN O I-OtherScientificTerm
are NN O O
most NN O O
powerful NN O O
and NN O O
identifies NN O O
a NN O O
number NN O O
of NN O O
linguistic NN O O
aspects NN O O
as NN O O
well NN O O
as NN O O
culture NN O O
- NN O O
and NN O O
domain-related NN O O
ones NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
investigated NN O O
whether NN O O
automatic NN O I-OtherScientificTerm
phonetic NN O I-OtherScientificTerm
transcriptions NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
APTs NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
can NN O O
replace NN O O
manually NN O I-OtherScientificTerm
verified NN O I-OtherScientificTerm
phonetic NN O I-OtherScientificTerm
transcriptions NN O I-OtherScientificTerm
-LRB- NN O O
MPTs NN O O
-RRB- NN O O
in NN O O
a NN O O
large NN O O
corpus-based NN O O
study NN O O
on NN O O
pronunciation NN O I-Task
variation NN O I-Task
. NN O O

To NN O O
this NN O O
end NN O O
, NN O O
we NN O O
compared NN O O
the NN O O
performance NN O O
of NN O O
both NN O O
transcription NN O O
types NN O O
in NN O O
a NN O O
classification NN O I-Task
experiment NN O O
aimed NN O O
at NN O O
establishing NN O O
the NN O O
direct NN O O
influence NN O O
of NN O O
a NN O O
particular NN O O
situational NN O O
setting NN O O
on NN O O
pronunciation NN O I-Task
variation NN O I-Task
. NN O O

We NN O O
trained NN O O
classifiers NN O I-Method
on NN O O
the NN O O
speech NN O I-Material
processes NN O I-Material
extracted NN O O
from NN O O
the NN O O
alignments NN O I-OtherScientificTerm
of NN O O
an NN O O
APT NN O I-OtherScientificTerm
and NN O O
an NN O O
MPT NN O I-OtherScientificTerm
with NN O O
a NN O O
canonical NN O I-OtherScientificTerm
transcription NN O I-OtherScientificTerm
. NN O O

We NN O O
tested NN O O
whether NN O O
the NN O O
classifiers NN O I-Material
were NN O O
equally NN O O
good NN O O
at NN O O
verifying NN O O
whether NN O O
unknown NN O I-OtherScientificTerm
transcriptions NN O I-OtherScientificTerm
represent NN O O
read NN O I-Material
speech NN O I-Material
or NN O O
telephone NN O I-Material
dialogues NN O I-Material
, NN O O
and NN O O
whether NN O O
the NN O O
same NN O O
speech NN O I-Material
processes NN O I-Material
were NN O O
identified NN O O
to NN O O
distinguish NN O O
between NN O O
transcriptions NN O O
of NN O O
the NN O O
two NN O O
situational NN O O
settings NN O O
. NN O O

Our NN O O
results NN O O
not NN O O
only NN O O
show NN O O
that NN O O
similar NN O O
distinguishing NN O O
speech NN O O
processes NN O O
were NN O O
identified NN O O
; NN O O
our NN O O
APT-based NN O I-Method
classifier NN O I-Method
yielded NN O O
better NN O O
classification NN O I-Metric
accuracy NN O I-Metric
than NN O O
the NN O O
MPT-based NN O I-Method
classifier NN O I-Method
whilst NN O O
using NN O O
fewer NN O O
classification NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Machine NN O I-Task
reading NN O I-Task
is NN O O
a NN O O
relatively NN O O
new NN O O
field NN O I-Generic
that NN O O
features NN O O
computer NN O I-Method
programs NN O I-Method
designed NN O O
to NN O O
read NN O O
flowing NN O I-Material
text NN O I-Material
and NN O O
extract NN O O
fact NN O I-OtherScientificTerm
assertions NN O I-OtherScientificTerm
expressed NN O O
by NN O O
the NN O O
narrative NN O I-Material
content NN O I-Material
. NN O O

This NN O O
task NN O I-Generic
involves NN O O
two NN O O
core NN O O
technologies NN O O
: NN O O
natural NN O I-Method
language NN O I-Method
processing NN O I-Method
-LRB- NN O I-Method
NLP NN O I-Method
-RRB- NN O I-Method
and NN O O
information NN O I-Task
extraction NN O I-Task
-LRB- NN O I-Task
IE NN O I-Task
-RRB- NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
describe NN O O
a NN O O
machine NN O I-Task
reading NN O I-Task
system NN O I-Task
that NN O O
we NN O O
have NN O O
developed NN O O
within NN O O
a NN O O
cognitive NN O I-Method
architecture NN O I-Method
. NN O O

We NN O O
show NN O O
how NN O O
we NN O O
have NN O O
integrated NN O O
into NN O O
the NN O O
framework NN O O
several NN O O
levels NN O O
of NN O O
knowledge NN O O
for NN O O
a NN O O
particular NN O O
domain NN O O
, NN O O
ideas NN O O
from NN O O
cognitive NN O I-OtherScientificTerm
semantics NN O I-OtherScientificTerm
and NN O O
construction NN O I-Method
grammar NN O I-Method
, NN O O
plus NN O O
tools NN O O
from NN O O
prior NN O I-Task
NLP NN O I-Task
and NN O O
IE NN O I-Task
research NN O I-Task
. NN O O

The NN O O
result NN O O
is NN O O
a NN O O
system NN O I-Generic
that NN O O
is NN O O
capable NN O O
of NN O O
reading NN O O
and NN O O
interpreting NN O O
complex NN O O
and NN O O
fairly NN O O
idiosyncratic NN O I-Material
texts NN O I-Material
in NN O O
the NN O O
family NN O I-Material
history NN O I-Material
domain NN O I-Material
. NN O O

We NN O O
describe NN O O
the NN O O
architecture NN O O
and NN O O
performance NN O O
of NN O O
the NN O O
system NN O I-Generic
. NN O O

After NN O O
presenting NN O O
the NN O O
results NN O O
from NN O O
several NN O O
evaluations NN O O
that NN O O
we NN O O
have NN O O
carried NN O O
out NN O O
, NN O O
we NN O O
summarize NN O O
possible NN O O
future NN O O
directions NN O O
. NN O O

-DOCSTART- NN O O

Nonstationary NN O I-OtherScientificTerm
chaotic NN O I-OtherScientificTerm
behavior NN O I-OtherScientificTerm
is NN O O
not NN O O
an NN O O
oxymoron NN O I-OtherScientificTerm
. NN O O

We NN O O
present NN O O
two NN O O
methods NN O I-Generic
for NN O O
capturing NN O O
nonstationary NN O I-OtherScientificTerm
chaos NN O I-OtherScientificTerm
, NN O O
then NN O O
present NN O O
a NN O O
few NN O O
examples NN O I-Generic
including NN O O
biological NN O I-Material
signals NN O I-Material
, NN O O
ocean NN O I-Material
waves NN O I-Material
and NN O O
traffic NN O I-Material
flow NN O I-Material
. NN O O

The NN O O
issue NN O O
is NN O O
of NN O O
practical NN O O
interest NN O O
because NN O O
it NN O O
is NN O O
often NN O O
useful NN O O
to NN O O
capture NN O O
when NN O O
nonstationary NN O I-OtherScientificTerm
events NN O I-OtherScientificTerm
take NN O O
place NN O O
and NN O O
it NN O O
is NN O O
desirable NN O O
to NN O O
know NN O O
over NN O O
what NN O O
periods NN O O
a NN O O
signal NN O O
is NN O O
stationary NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
formal NN O I-Method
analysis NN O I-Method
for NN O O
a NN O O
large NN O O
class NN O O
of NN O O
words NN O O
called NN O O
alternative NN O I-OtherScientificTerm
markers NN O I-OtherScientificTerm
, NN O O
which NN O O
includes NN O O
other NN O O
-LRB- NN O O
than NN O O
-RRB- NN O O
, NN O O
such NN O O
-LRB- NN O O
as NN O O
-RRB- NN O O
, NN O O
and NN O O
besides NN O O
. NN O O

These NN O O
words NN O I-Generic
appear NN O O
frequently NN O O
enough NN O O
in NN O O
dialog NN O I-OtherScientificTerm
to NN O O
warrant NN O O
serious NN O O
attention NN O O
, NN O O
yet NN O O
present NN O O
natural NN O I-Method
language NN O I-Method
search NN O I-Method
engines NN O I-Method
perform NN O O
poorly NN O O
on NN O O
queries NN O O
containing NN O O
them NN O I-Generic
. NN O O

I NN O O
show NN O O
that NN O O
the NN O O
performance NN O O
of NN O O
a NN O O
search NN O I-Method
engine NN O I-Method
can NN O O
be NN O O
improved NN O O
dramatically NN O O
by NN O O
incorporating NN O O
an NN O O
approximation NN O I-Method
of NN O I-Method
the NN O I-Method
formal NN O I-Method
analysis NN O I-Method
that NN O O
is NN O O
compatible NN O O
with NN O O
the NN O O
search NN O I-Method
engine NN O I-Method
's NN O O
operational NN O I-OtherScientificTerm
semantics NN O I-OtherScientificTerm
. NN O O

The NN O O
value NN O O
of NN O O
this NN O O
approach NN O I-Generic
is NN O O
that NN O O
as NN O O
the NN O O
operational NN O I-OtherScientificTerm
semantics NN O I-OtherScientificTerm
of NN O O
natural NN O I-Task
language NN O I-Task
applications NN O I-Task
improve NN O O
, NN O O
even NN O O
larger NN O O
improvements NN O O
are NN O O
possible NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
address NN O O
the NN O O
problem NN O O
of NN O O
combining NN O O
several NN O O
language NN O I-Method
models NN O I-Method
-LRB- NN O I-Method
LMs NN O I-Method
-RRB- NN O I-Method
. NN O O

We NN O O
find NN O O
that NN O O
simple NN O O
interpolation NN O I-Method
methods NN O I-Method
, NN O O
like NN O O
log-linear NN O I-Method
and NN O I-Method
linear NN O I-Method
interpolation NN O I-Method
, NN O O
improve NN O O
the NN O O
performance NN O O
but NN O O
fall NN O O
short NN O O
of NN O O
the NN O O
performance NN O O
of NN O O
an NN O O
oracle NN O O
. NN O O

The NN O O
oracle NN O O
knows NN O O
the NN O O
reference NN O O
word NN O O
string NN O O
and NN O O
selects NN O O
the NN O O
word NN O O
string NN O O
with NN O O
the NN O O
best NN O O
performance NN O O
-LRB- NN O O
typically NN O O
, NN O O
word NN O I-Metric
or NN O I-Metric
semantic NN O I-Metric
error NN O I-Metric
rate NN O I-Metric
-RRB- NN O O
from NN O O
a NN O O
list NN O O
of NN O O
word NN O O
strings NN O O
, NN O O
where NN O O
each NN O O
word NN O O
string NN O O
has NN O O
been NN O O
obtained NN O O
by NN O O
using NN O O
a NN O O
different NN O O
LM NN O I-Method
. NN O O

Actually NN O O
, NN O O
the NN O O
oracle NN O O
acts NN O O
like NN O O
a NN O O
dynamic NN O I-Method
combiner NN O I-Method
with NN O O
hard NN O I-OtherScientificTerm
decisions NN O I-OtherScientificTerm
using NN O O
the NN O O
reference NN O O
. NN O O

We NN O O
provide NN O O
experimental NN O O
results NN O O
that NN O O
clearly NN O O
show NN O O
the NN O O
need NN O O
for NN O O
a NN O O
dynamic NN O I-Method
language NN O I-Method
model NN O I-Method
combination NN O I-Method
to NN O O
improve NN O O
the NN O O
performance NN O O
further NN O O
. NN O O

We NN O O
suggest NN O O
a NN O O
method NN O I-Generic
that NN O O
mimics NN O O
the NN O O
behavior NN O O
of NN O O
the NN O O
oracle NN O O
using NN O O
a NN O O
neural NN O I-Method
network NN O I-Method
or NN O O
a NN O O
decision NN O I-Method
tree NN O I-Method
. NN O O

The NN O O
method NN O I-Generic
amounts NN O O
to NN O O
tagging NN O O
LMs NN O I-Method
with NN O O
confidence NN O I-Metric
measures NN O I-Metric
and NN O O
picking NN O O
the NN O O
best NN O O
hypothesis NN O O
corresponding NN O O
to NN O O
the NN O O
LM NN O I-Method
with NN O O
the NN O O
best NN O O
confidence NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
a NN O O
new NN O O
method NN O I-Generic
for NN O O
the NN O O
representation NN O O
of NN O O
NLP NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
within NN O O
reranking NN O I-Method
approaches NN O I-Method
. NN O O

We NN O O
make NN O O
use NN O O
of NN O O
a NN O O
conditional NN O I-Method
log-linear NN O I-Method
model NN O I-Method
, NN O O
with NN O O
hidden NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
representing NN O O
the NN O O
assignment NN O O
of NN O O
lexical NN O O
items NN O O
to NN O O
word NN O I-OtherScientificTerm
clusters NN O I-OtherScientificTerm
or NN O O
word NN O I-OtherScientificTerm
senses NN O I-OtherScientificTerm
. NN O O

The NN O O
model NN O I-Generic
learns NN O O
to NN O O
automatically NN O O
make NN O O
these NN O O
assignments NN O O
based NN O O
on NN O O
a NN O O
discriminative NN O I-Metric
training NN O I-Metric
criterion NN O I-Metric
. NN O O

Training NN O O
and NN O O
decoding NN O O
with NN O O
the NN O O
model NN O I-Generic
requires NN O O
summing NN O O
over NN O O
an NN O O
exponential NN O O
number NN O O
of NN O O
hidden-variable NN O I-OtherScientificTerm
assignments NN O I-OtherScientificTerm
: NN O O
the NN O O
required NN O O
summations NN O I-Generic
can NN O O
be NN O O
computed NN O O
efficiently NN O O
and NN O O
exactly NN O O
using NN O O
dynamic NN O I-Method
programming NN O I-Method
. NN O O

As NN O O
a NN O O
case NN O O
study NN O O
, NN O O
we NN O O
apply NN O O
the NN O O
model NN O I-Generic
to NN O O
parse NN O I-Task
reranking NN O I-Task
. NN O O

The NN O O
model NN O I-Generic
gives NN O O
an NN O O
F-measure NN O I-Metric
improvement NN O O
of NN O O
~ NN O O
1.25 NN O O
% NN O O
beyond NN O O
the NN O O
base NN O I-Method
parser NN O I-Method
, NN O O
and NN O O
an NN O O
~ NN O O
0.25 NN O O
% NN O O
improvement NN O O
beyond NN O O
Collins NN O I-Method
-LRB- NN O I-Method
2000 NN O I-Method
-RRB- NN O I-Method
reranker NN O I-Method
. NN O O

Although NN O O
our NN O O
experiments NN O O
are NN O O
focused NN O O
on NN O O
parsing NN O I-Task
, NN O O
the NN O O
techniques NN O I-Generic
described NN O O
generalize NN O O
naturally NN O O
to NN O O
NLP NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
other NN O O
than NN O O
parse NN O I-OtherScientificTerm
trees NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
an NN O O
algorithm NN O I-Generic
for NN O O
learning NN O I-Task
the NN O I-Task
time-varying NN O I-Task
shape NN O I-Task
of NN O I-Task
a NN O I-Task
non-rigid NN O I-Task
3D NN O I-Task
object NN O I-Task
from NN O O
uncalibrated NN O I-Material
2D NN O I-Material
tracking NN O I-Material
data NN O I-Material
. NN O O

We NN O O
model NN O O
shape NN O I-OtherScientificTerm
motion NN O I-OtherScientificTerm
as NN O O
a NN O O
rigid NN O I-Method
component NN O I-Method
-LRB- NN O O
rotation NN O I-OtherScientificTerm
and NN O O
translation NN O I-OtherScientificTerm
-RRB- NN O O
combined NN O O
with NN O O
a NN O O
non-rigid NN O I-OtherScientificTerm
deformation NN O I-OtherScientificTerm
. NN O O

Reconstruction NN O I-OtherScientificTerm
is NN O O
ill-posed NN O O
if NN O O
arbitrary NN O I-OtherScientificTerm
deformations NN O I-OtherScientificTerm
are NN O O
allowed NN O O
. NN O O

We NN O O
constrain NN O O
the NN O O
problem NN O O
by NN O O
assuming NN O O
that NN O O
the NN O O
object NN O I-OtherScientificTerm
shape NN O I-OtherScientificTerm
at NN O O
each NN O O
time NN O O
instant NN O O
is NN O O
drawn NN O O
from NN O O
a NN O O
Gaussian NN O I-Method
distribution NN O I-Method
. NN O O

Based NN O O
on NN O O
this NN O O
assumption NN O O
, NN O O
the NN O O
algorithm NN O I-Generic
simultaneously NN O O
estimates NN O O
3D NN O I-OtherScientificTerm
shape NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
motion NN O I-OtherScientificTerm
for NN O O
each NN O O
time NN O O
frame NN O O
, NN O O
learns NN O O
the NN O O
parameters NN O O
of NN O O
the NN O O
Gaussian NN O I-Method
, NN O O
and NN O O
robustly NN O O
fills-in NN O O
missing NN O O
data NN O O
points NN O O
. NN O O

We NN O O
then NN O O
extend NN O O
the NN O O
algorithm NN O I-Generic
to NN O O
model NN O O
temporal NN O I-OtherScientificTerm
smoothness NN O I-OtherScientificTerm
in NN O I-OtherScientificTerm
object NN O I-OtherScientificTerm
shape NN O I-OtherScientificTerm
, NN O O
thus NN O O
allowing NN O O
it NN O I-Generic
to NN O O
handle NN O O
severe NN O O
cases NN O O
of NN O O
missing NN O I-Material
data NN O I-Material
. NN O O

-DOCSTART- NN O O

Automatic NN O I-Task
summarization NN O I-Task
and NN O O
information NN O I-Task
extraction NN O I-Task
are NN O O
two NN O O
important NN O O
Internet NN O O
services NN O O
. NN O O

MUC NN O I-Material
and NN O O
SUMMAC NN O I-Material
play NN O O
their NN O O
appropriate NN O O
roles NN O O
in NN O O
the NN O O
next NN O O
generation NN O O
Internet NN O O
. NN O O

This NN O O
paper NN O O
focuses NN O O
on NN O O
the NN O O
automatic NN O I-Task
summarization NN O I-Task
and NN O O
proposes NN O O
two NN O O
different NN O O
models NN O I-Generic
to NN O O
extract NN O O
sentences NN O O
for NN O O
summary NN O I-Task
generation NN O I-Task
under NN O O
two NN O O
tasks NN O I-Generic
initiated NN O O
by NN O O
SUMMAC-1 NN O I-Material
. NN O O

For NN O O
categorization NN O I-Task
task NN O I-Task
, NN O O
positive NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
vectors NN O I-OtherScientificTerm
and NN O O
negative NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
vectors NN O I-OtherScientificTerm
are NN O O
used NN O O
cooperatively NN O O
to NN O O
construct NN O O
generic NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
indicative NN O I-OtherScientificTerm
summaries NN O I-OtherScientificTerm
. NN O O

For NN O O
adhoc NN O I-Task
task NN O I-Task
, NN O O
a NN O O
text NN O I-Method
model NN O I-Method
based NN O O
on NN O O
relationship NN O O
between NN O O
nouns NN O O
and NN O O
verbs NN O O
is NN O O
used NN O O
to NN O O
filter NN O O
out NN O O
irrelevant NN O O
discourse NN O I-OtherScientificTerm
segment NN O I-OtherScientificTerm
, NN O O
to NN O O
rank NN O O
relevant NN O O
sentences NN O O
, NN O O
and NN O O
to NN O O
generate NN O O
the NN O O
user-directed NN O I-OtherScientificTerm
summaries NN O I-OtherScientificTerm
. NN O O

The NN O O
result NN O O
shows NN O O
that NN O O
the NN O O
NormF NN O I-Metric
of NN O O
the NN O O
best NN O O
summary NN O O
and NN O O
that NN O O
of NN O O
the NN O O
fixed NN O O
summary NN O O
for NN O O
adhoc NN O I-Task
tasks NN O I-Task
are NN O O
0.456 NN O O
and NN O O
0 NN O O
. NN O O

447 NN O O
. NN O O

The NN O O
NormF NN O I-Metric
of NN O O
the NN O O
best NN O O
summary NN O O
and NN O O
that NN O O
of NN O O
the NN O O
fixed NN O O
summary NN O O
for NN O O
categorization NN O I-Task
task NN O I-Task
are NN O O
0.4090 NN O O
and NN O O
0.4023 NN O O
. NN O O

Our NN O O
system NN O I-Generic
outperforms NN O O
the NN O O
average NN O O
system NN O I-Generic
in NN O O
categorization NN O I-Task
task NN O I-Task
but NN O O
does NN O O
a NN O O
common NN O O
job NN O O
in NN O O
adhoc NN O I-Task
task NN O I-Task
. NN O O

-DOCSTART- NN O O

In NN O O
real-world NN O I-Task
action NN O I-Task
recognition NN O I-Task
problems NN O I-Task
, NN O O
low-level NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
can NN O O
not NN O O
adequately NN O O
characterize NN O O
the NN O O
rich NN O I-OtherScientificTerm
spatial-temporal NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
in NN O O
action NN O I-Material
videos NN O I-Material
. NN O O

In NN O O
this NN O O
work NN O O
, NN O O
we NN O O
encode NN O O
actions NN O O
based NN O O
on NN O O
attributes NN O O
that NN O O
describes NN O O
actions NN O O
as NN O O
high-level NN O I-OtherScientificTerm
concepts NN O I-OtherScientificTerm
e.g. NN O O
, NN O O
jump NN O O
forward NN O O
or NN O O
motion NN O O
in NN O O
the NN O O
air NN O O
. NN O O

We NN O O
base NN O O
our NN O O
analysis NN O O
on NN O O
two NN O O
types NN O O
of NN O O
action NN O I-OtherScientificTerm
attributes NN O I-OtherScientificTerm
. NN O O

One NN O O
type NN O O
of NN O O
action NN O I-OtherScientificTerm
attributes NN O I-OtherScientificTerm
is NN O O
generated NN O O
by NN O O
humans NN O O
. NN O O

The NN O O
second NN O O
type NN O O
is NN O O
data-driven NN O I-OtherScientificTerm
attributes NN O I-OtherScientificTerm
, NN O O
which NN O O
are NN O O
learned NN O O
from NN O O
data NN O O
using NN O O
dictionary NN O I-Method
learning NN O I-Method
methods NN O I-Method
. NN O O

Attribute-based NN O I-Method
representation NN O I-Method
may NN O O
exhibit NN O O
high NN O O
variance NN O O
due NN O O
to NN O O
noisy NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
redundant NN O I-OtherScientificTerm
attributes NN O I-OtherScientificTerm
. NN O O

We NN O O
propose NN O O
a NN O O
discriminative NN O I-Method
and NN O I-Method
compact NN O I-Method
attribute-based NN O I-Method
representation NN O I-Method
by NN O O
selecting NN O O
a NN O O
subset NN O O
of NN O O
discriminative NN O I-OtherScientificTerm
attributes NN O I-OtherScientificTerm
from NN O O
a NN O O
large NN O O
attribute NN O O
set NN O O
. NN O O

Three NN O O
attribute NN O I-Metric
selection NN O I-Metric
criteria NN O I-Metric
are NN O O
proposed NN O O
and NN O O
formulated NN O O
as NN O O
a NN O O
submodular NN O I-Task
optimization NN O I-Task
problem NN O I-Task
. NN O O

A NN O O
greedy NN O I-Method
optimization NN O I-Method
algorithm NN O I-Method
is NN O O
presented NN O O
and NN O O
guaranteed NN O O
to NN O O
be NN O O
at NN O O
least NN O O
-LRB- NN O O
1-1 NN O O
/ NN O O
e NN O O
-RRB- NN O O
- NN O O
approximation NN O O
to NN O O
the NN O O
optimum NN O O
. NN O O

Experimental NN O O
results NN O O
on NN O O
the NN O O
Olympic NN O I-Material
Sports NN O I-Material
and NN O I-Material
UCF101 NN O I-Material
datasets NN O I-Material
demonstrate NN O O
that NN O O
the NN O O
proposed NN O O
attribute-based NN O I-Method
representation NN O I-Method
can NN O O
significantly NN O O
boost NN O O
the NN O O
performance NN O O
of NN O O
action NN O I-Method
recognition NN O I-Method
algorithms NN O I-Method
and NN O O
outperform NN O O
most NN O O
recently NN O O
proposed NN O O
recognition NN O I-Method
approaches NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
reports NN O O
a NN O O
completed NN O O
stage NN O O
of NN O O
ongoing NN O O
research NN O O
at NN O O
the NN O O
University NN O O
of NN O O
York NN O O
. NN O O

Landsbergen NN O O
's NN O O
advocacy NN O O
of NN O O
analytical NN O I-Method
inverses NN O I-Method
for NN O O
compositional NN O I-OtherScientificTerm
syntax NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
encourages NN O O
the NN O O
application NN O O
of NN O O
Definite NN O I-Method
Clause NN O I-Method
Grammar NN O I-Method
techniques NN O I-Method
to NN O O
the NN O O
construction NN O O
of NN O O
a NN O O
parser NN O I-Task
returning NN O I-Task
Montague NN O I-Task
analysis NN O I-Task
trees NN O I-Task
. NN O O

A NN O O
parser NN O I-Method
MDCC NN O I-Method
is NN O O
presented NN O O
which NN O O
implements NN O O
an NN O O
augmented NN O I-Method
Friedman NN O I-Method
- NN O I-Method
Warren NN O I-Method
algorithm NN O I-Method
permitting NN O O
post NN O I-OtherScientificTerm
referencing NN O I-OtherScientificTerm
* NN O O
and NN O O
interfaces NN O O
with NN O O
a NN O O
language NN O O
of NN O O
intenslonal NN O I-OtherScientificTerm
logic NN O I-OtherScientificTerm
translator NN O I-OtherScientificTerm
LILT NN O I-OtherScientificTerm
so NN O O
as NN O O
to NN O O
display NN O O
the NN O O
derivational NN O I-OtherScientificTerm
history NN O I-OtherScientificTerm
of NN O O
corresponding NN O O
reduced NN O I-OtherScientificTerm
IL NN O I-OtherScientificTerm
formulae NN O I-OtherScientificTerm
. NN O O

Some NN O O
familiarity NN O O
with NN O O
Montague NN O I-Method
's NN O I-Method
PTQ NN O I-Method
and NN O O
the NN O O
basic NN O I-Method
DCG NN O I-Method
mechanism NN O I-Method
is NN O O
assumed NN O O
. NN O O

-DOCSTART- NN O O

Despite NN O O
their NN O O
success NN O O
, NN O O
convolutional NN O I-Method
neural NN O I-Method
networks NN O I-Method
are NN O O
computationally NN O O
expensive NN O O
because NN O O
they NN O I-Generic
must NN O O
examine NN O O
all NN O O
image NN O O
locations NN O O
. NN O O

Stochastic NN O I-Method
attention-based NN O I-Method
models NN O I-Method
have NN O O
been NN O O
shown NN O O
to NN O O
improve NN O O
computational NN O I-Metric
efficiency NN O I-Metric
at NN O O
test NN O O
time NN O O
, NN O O
but NN O O
they NN O I-Generic
remain NN O O
difficult NN O O
to NN O O
train NN O O
because NN O O
of NN O O
intractable NN O I-Task
posterior NN O I-Task
inference NN O I-Task
and NN O O
high NN O O
variance NN O O
in NN O O
the NN O O
stochastic NN O I-Task
gradient NN O I-Task
estimates NN O I-Task
. NN O O

Borrowing NN O I-Method
techniques NN O I-Method
from NN O O
the NN O O
literature NN O O
on NN O O
training NN O O
deep NN O I-Method
generative NN O I-Method
models NN O I-Method
, NN O O
we NN O O
present NN O O
the NN O O
Wake-Sleep NN O I-Method
Recurrent NN O I-Method
Attention NN O I-Method
Model NN O I-Method
, NN O O
a NN O O
method NN O I-Generic
for NN O O
training NN O O
stochastic NN O I-Method
attention NN O I-Method
networks NN O I-Method
which NN O O
improves NN O O
posterior NN O I-Task
inference NN O I-Task
and NN O O
which NN O O
reduces NN O O
the NN O O
variability NN O O
in NN O O
the NN O O
stochastic NN O I-OtherScientificTerm
gradients NN O I-OtherScientificTerm
. NN O O

We NN O O
show NN O O
that NN O O
our NN O O
method NN O I-Generic
can NN O O
greatly NN O O
speed NN O O
up NN O O
the NN O O
training NN O I-Metric
time NN O I-Metric
for NN O O
stochastic NN O I-Method
attention NN O I-Method
networks NN O I-Method
in NN O O
the NN O O
domains NN O O
of NN O O
image NN O I-Task
classification NN O I-Task
and NN O O
caption NN O I-Task
generation NN O I-Task
. NN O O

-DOCSTART- NN O O

A NN O O
new NN O O
exemplar-based NN O I-Method
framework NN O I-Method
unifying NN O O
image NN O I-Task
completion NN O I-Task
, NN O O
texture NN O I-Task
synthesis NN O I-Task
and NN O O
image NN O I-Task
inpainting NN O I-Task
is NN O O
presented NN O O
in NN O O
this NN O O
work NN O O
. NN O O

Contrary NN O O
to NN O O
existing NN O O
greedy NN O I-Method
techniques NN O I-Method
, NN O O
these NN O O
tasks NN O I-Generic
are NN O O
posed NN O O
in NN O O
the NN O O
form NN O O
of NN O O
a NN O O
discrete NN O I-Task
global NN O I-Task
optimization NN O I-Task
problem NN O I-Task
with NN O O
a NN O O
well NN O I-OtherScientificTerm
defined NN O I-OtherScientificTerm
objective NN O I-OtherScientificTerm
function NN O I-OtherScientificTerm
. NN O O

For NN O O
solving NN O O
this NN O O
problem NN O I-Generic
a NN O O
novel NN O O
optimization NN O I-Method
scheme NN O I-Method
, NN O O
called NN O O
Priority-BP NN O I-Method
, NN O O
is NN O O
proposed NN O O
which NN O O
carries NN O O
two NN O O
very NN O O
important NN O O
extensions NN O I-Generic
over NN O O
standard NN O O
belief NN O I-Method
propagation NN O I-Method
-LRB- NN O I-Method
BP NN O I-Method
-RRB- NN O I-Method
: NN O O
'' NN O O
priority-based NN O I-Method
message NN O I-Method
scheduling NN O I-Method
'' NN O O
and NN O O
'' NN O O
dynamic NN O I-Method
label NN O I-Method
pruning NN O I-Method
'' NN O O
. NN O O

These NN O O
two NN O O
extensions NN O I-Generic
work NN O O
in NN O O
cooperation NN O O
to NN O O
deal NN O O
with NN O O
the NN O O
intolerable NN O I-OtherScientificTerm
computational NN O I-OtherScientificTerm
cost NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
BP NN O I-Method
caused NN O O
by NN O O
the NN O O
huge NN O O
number NN O O
of NN O O
existing NN O O
labels NN O O
. NN O O

Moreover NN O O
, NN O O
both NN O O
extensions NN O I-Generic
are NN O O
generic NN O O
and NN O O
can NN O O
therefore NN O O
be NN O O
applied NN O O
to NN O O
any NN O O
MRF NN O I-Method
energy NN O I-Method
function NN O I-Method
as NN O O
well NN O O
. NN O O

The NN O O
effectiveness NN O O
of NN O O
our NN O O
method NN O I-Generic
is NN O O
demonstrated NN O O
on NN O O
a NN O O
wide NN O O
variety NN O O
of NN O O
image NN O I-Material
completion NN O I-Material
examples NN O I-Material
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
compare NN O O
the NN O O
relative NN O O
effects NN O O
of NN O O
segment NN O I-OtherScientificTerm
order NN O I-OtherScientificTerm
, NN O O
segmentation NN O I-OtherScientificTerm
and NN O O
segment NN O I-OtherScientificTerm
contiguity NN O I-OtherScientificTerm
on NN O O
the NN O O
retrieval NN O I-Task
performance NN O O
of NN O O
a NN O O
translation NN O I-Method
memory NN O I-Method
system NN O I-Method
. NN O O

We NN O O
take NN O O
a NN O O
selection NN O O
of NN O O
both NN O O
bag-of-words NN O I-Method
and NN O I-Method
segment NN O I-Method
order-sensitive NN O I-Method
string NN O I-Method
comparison NN O I-Method
methods NN O I-Method
, NN O O
and NN O O
run NN O O
each NN O O
over NN O O
both NN O O
character NN O I-Material
- NN O I-Material
and NN O I-Material
word-segmented NN O I-Material
data NN O I-Material
, NN O O
in NN O O
combination NN O O
with NN O O
a NN O O
range NN O O
of NN O O
local NN O I-Method
segment NN O I-Method
contiguity NN O I-Method
models NN O I-Method
-LRB- NN O O
in NN O O
the NN O O
form NN O O
of NN O O
N-grams NN O I-Method
-RRB- NN O O
. NN O O

Over NN O O
two NN O O
distinct NN O O
datasets NN O O
, NN O O
we NN O O
find NN O O
that NN O O
indexing NN O I-Task
according NN O O
to NN O O
simple NN O O
character NN O I-Method
bigrams NN O I-Method
produces NN O O
a NN O O
retrieval NN O I-Metric
accuracy NN O I-Metric
superior NN O O
to NN O O
any NN O O
of NN O O
the NN O O
tested NN O O
word NN O I-Method
N-gram NN O I-Method
models NN O I-Method
. NN O O

Further NN O O
, NN O O
in NN O O
their NN O O
optimum NN O O
configuration NN O O
, NN O O
bag-of-words NN O I-Method
methods NN O I-Method
are NN O O
shown NN O O
to NN O O
be NN O O
equivalent NN O O
to NN O O
segment NN O I-Method
order-sensitive NN O I-Method
methods NN O I-Method
in NN O O
terms NN O O
of NN O O
retrieval NN O I-Metric
accuracy NN O I-Metric
, NN O O
but NN O O
much NN O O
faster NN O O
. NN O O

We NN O O
also NN O O
provide NN O O
evidence NN O O
that NN O O
our NN O O
findings NN O O
are NN O O
scalable NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
show NN O O
how NN O O
two NN O O
standard NN O O
outputs NN O I-Generic
from NN O O
information NN O I-Method
extraction NN O I-Method
-LRB- NN O I-Method
IE NN O I-Method
-RRB- NN O I-Method
systems NN O I-Method
- NN O O
named NN O I-OtherScientificTerm
entity NN O I-OtherScientificTerm
annotations NN O I-OtherScientificTerm
and NN O O
scenario NN O I-OtherScientificTerm
templates NN O I-OtherScientificTerm
- NN O O
can NN O O
be NN O O
used NN O O
to NN O O
enhance NN O O
access NN O O
to NN O O
text NN O I-Material
collections NN O I-Material
via NN O O
a NN O O
standard NN O O
text NN O I-Method
browser NN O I-Method
. NN O O

We NN O O
describe NN O O
how NN O O
this NN O O
information NN O O
is NN O O
used NN O O
in NN O O
a NN O O
prototype NN O I-Generic
system NN O I-Generic
designed NN O O
to NN O O
support NN O O
information NN O O
workers NN O O
' NN O O
access NN O O
to NN O O
a NN O O
pharmaceutical NN O I-Material
news NN O I-Material
archive NN O I-Material
as NN O O
part NN O O
of NN O O
their NN O O
industry NN O O
watch NN O O
function NN O O
. NN O O

We NN O O
also NN O O
report NN O O
results NN O O
of NN O O
a NN O O
preliminary NN O O
, NN O O
qualitative NN O I-Metric
user NN O I-Metric
evaluation NN O I-Metric
of NN O O
the NN O O
system NN O I-Generic
, NN O O
which NN O O
while NN O O
broadly NN O O
positive NN O O
indicates NN O O
further NN O O
work NN O O
needs NN O O
to NN O O
be NN O O
done NN O O
on NN O O
the NN O O
interface NN O O
to NN O O
make NN O O
users NN O O
aware NN O O
of NN O O
the NN O O
increased NN O O
potential NN O O
of NN O O
IE-enhanced NN O I-Method
text NN O I-Method
browsers NN O I-Method
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
a NN O O
new NN O O
model-based NN O I-Method
bundle NN O I-Method
adjustment NN O I-Method
algorithm NN O I-Method
to NN O O
recover NN O O
the NN O O
3D NN O I-Method
model NN O I-Method
of NN O O
a NN O O
scene/object NN O O
from NN O O
a NN O O
sequence NN O O
of NN O O
images NN O I-Material
with NN O O
unknown NN O I-OtherScientificTerm
motions NN O I-OtherScientificTerm
. NN O O

Instead NN O O
of NN O O
representing NN O O
scene/object NN O O
by NN O O
a NN O O
collection NN O O
of NN O O
isolated NN O I-OtherScientificTerm
3D NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
-LRB- NN O O
usually NN O O
points NN O O
-RRB- NN O O
, NN O O
our NN O O
algorithm NN O I-Generic
uses NN O O
a NN O O
surface NN O I-OtherScientificTerm
controlled NN O O
by NN O O
a NN O O
small NN O O
set NN O O
of NN O O
parameters NN O O
. NN O O

Compared NN O O
with NN O O
previous NN O O
model-based NN O I-Generic
approaches NN O I-Generic
, NN O O
our NN O O
approach NN O I-Generic
has NN O O
the NN O O
following NN O O
advantages NN O O
. NN O O

First NN O O
, NN O O
instead NN O O
of NN O O
using NN O O
the NN O O
model NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
as NN O O
a NN O O
regular-izer NN O I-OtherScientificTerm
, NN O O
we NN O O
directly NN O O
use NN O O
it NN O I-Generic
as NN O O
our NN O O
search NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
, NN O O
thus NN O O
resulting NN O O
in NN O O
a NN O O
more NN O O
elegant NN O O
formulation NN O O
with NN O O
fewer NN O O
unknowns NN O O
and NN O O
fewer NN O O
equations NN O O
. NN O O

Second NN O O
, NN O O
our NN O O
algorithm NN O I-Generic
automatically NN O O
associates NN O O
tracked NN O O
points NN O O
with NN O O
their NN O O
correct NN O O
locations NN O O
on NN O O
the NN O O
surfaces NN O I-OtherScientificTerm
, NN O O
thereby NN O O
eliminating NN O O
the NN O O
need NN O O
for NN O O
a NN O O
prior NN O I-OtherScientificTerm
2D-to-3D NN O I-OtherScientificTerm
association NN O I-OtherScientificTerm
. NN O O

Third NN O O
, NN O O
regarding NN O O
face NN O I-Task
modeling NN O I-Task
, NN O O
we NN O O
use NN O O
a NN O O
very NN O O
small NN O O
set NN O O
of NN O O
face NN O I-Metric
metrics NN O I-Metric
-LRB- NN O O
meaningful NN O O
deformations NN O O
-RRB- NN O O
to NN O O
parame-terize NN O O
the NN O O
face NN O I-OtherScientificTerm
geometry NN O I-OtherScientificTerm
, NN O O
resulting NN O O
in NN O O
a NN O O
smaller NN O O
search NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
and NN O O
a NN O O
better NN O O
posed NN O I-Method
system NN O I-Method
. NN O O

Experiments NN O O
with NN O O
both NN O O
synthetic NN O I-Material
and NN O I-Material
real NN O I-Material
data NN O I-Material
show NN O O
that NN O O
this NN O O
new NN O O
algorithm NN O I-Generic
is NN O O
faster NN O O
, NN O O
more NN O O
accurate NN O O
and NN O O
more NN O O
stable NN O O
than NN O O
existing NN O O
ones NN O I-Generic
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
an NN O O
approach NN O I-Generic
to NN O O
the NN O O
unsupervised NN O I-Task
learning NN O I-Task
of NN O I-Task
parts NN O I-Task
of NN O I-Task
speech NN O I-Task
which NN O O
uses NN O O
both NN O O
morphological NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
syntactic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
. NN O O

While NN O O
the NN O O
model NN O I-Generic
is NN O O
more NN O O
complex NN O O
than NN O O
those NN O I-Generic
which NN O O
have NN O O
been NN O O
employed NN O O
for NN O O
unsupervised NN O I-Task
learning NN O I-Task
of NN O I-Task
POS NN O I-Task
tags NN O I-Task
in NN O I-Task
English NN O I-Task
, NN O O
which NN O O
use NN O O
only NN O O
syntactic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
, NN O O
the NN O O
variety NN O O
of NN O O
languages NN O O
in NN O O
the NN O O
world NN O O
requires NN O O
that NN O O
we NN O O
consider NN O O
morphology NN O I-OtherScientificTerm
as NN O O
well NN O O
. NN O O

In NN O O
many NN O O
languages NN O O
, NN O O
morphology NN O I-OtherScientificTerm
provides NN O O
better NN O O
clues NN O O
to NN O O
a NN O O
word NN O O
's NN O O
category NN O O
than NN O O
word NN O I-OtherScientificTerm
order NN O I-OtherScientificTerm
. NN O O

We NN O O
present NN O O
the NN O O
computational NN O I-Generic
model NN O I-Generic
for NN O O
POS NN O I-Task
learning NN O I-Task
, NN O O
and NN O O
present NN O O
results NN O O
for NN O O
applying NN O O
it NN O I-Generic
to NN O O
Bulgarian NN O I-Material
, NN O O
a NN O O
Slavic NN O I-Material
language NN O I-Material
with NN O O
relatively NN O O
free NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
order NN O I-OtherScientificTerm
and NN O O
rich NN O I-OtherScientificTerm
morphology NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Words NN O O
in NN O O
Chinese NN O I-Material
text NN O I-Material
are NN O O
not NN O O
naturally NN O O
separated NN O O
by NN O O
delimiters NN O O
, NN O O
which NN O O
poses NN O O
a NN O O
challenge NN O O
to NN O O
standard NN O O
machine NN O I-Task
translation NN O I-Task
-LRB- NN O I-Task
MT NN O I-Task
-RRB- NN O I-Task
systems NN O I-Task
. NN O O

In NN O O
MT NN O I-Task
, NN O O
the NN O O
widely NN O O
used NN O O
approach NN O O
is NN O O
to NN O O
apply NN O O
a NN O O
Chinese NN O I-Method
word NN O I-Method
segmenter NN O I-Method
trained NN O O
from NN O O
manually NN O I-Material
annotated NN O I-Material
data NN O I-Material
, NN O O
using NN O O
a NN O O
fixed NN O O
lexicon NN O O
. NN O O

Such NN O O
word NN O I-Method
segmentation NN O I-Method
is NN O O
not NN O O
necessarily NN O O
optimal NN O O
for NN O O
translation NN O I-Task
. NN O O

We NN O O
propose NN O O
a NN O O
Bayesian NN O I-Method
semi-supervised NN O I-Method
Chinese NN O I-Method
word NN O I-Method
segmentation NN O I-Method
model NN O I-Method
which NN O O
uses NN O O
both NN O O
monolingual NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
bilingual NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
to NN O O
derive NN O O
a NN O O
segmentation NN O I-Task
suitable NN O O
for NN O O
MT NN O I-Task
. NN O O

Experiments NN O O
show NN O O
that NN O O
our NN O O
method NN O I-Generic
improves NN O O
a NN O O
state-of-the-art NN O O
MT NN O I-Method
system NN O I-Method
in NN O O
a NN O O
small NN O O
and NN O O
a NN O O
large NN O O
data NN O O
environment NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
compare NN O O
two NN O O
competing NN O O
approaches NN O I-Generic
to NN O O
part-of-speech NN O I-Task
tagging NN O I-Task
, NN O O
statistical NN O I-Task
and NN O I-Task
constraint-based NN O I-Task
disambiguation NN O I-Task
, NN O O
using NN O O
French NN O I-Material
as NN O O
our NN O O
test NN O O
language NN O O
. NN O O

We NN O O
imposed NN O O
a NN O O
time NN O O
limit NN O O
on NN O O
our NN O O
experiment NN O O
: NN O O
the NN O O
amount NN O O
of NN O O
time NN O O
spent NN O O
on NN O O
the NN O O
design NN O O
of NN O O
our NN O O
constraint NN O I-Generic
system NN O I-Generic
was NN O O
about NN O O
the NN O O
same NN O O
as NN O O
the NN O O
time NN O O
we NN O O
used NN O O
to NN O O
train NN O O
and NN O O
test NN O O
the NN O O
easy-to-implement NN O O
statistical NN O I-Method
model NN O I-Method
. NN O O

We NN O O
describe NN O O
the NN O O
two NN O O
systems NN O I-Generic
and NN O O
compare NN O O
the NN O O
results NN O O
. NN O O

The NN O O
accuracy NN O I-Metric
of NN O O
the NN O O
statistical NN O I-Method
method NN O I-Method
is NN O O
reasonably NN O O
good NN O O
, NN O O
comparable NN O O
to NN O O
taggers NN O I-Method
for NN O O
English NN O I-Material
. NN O O

But NN O O
the NN O O
constraint-based NN O I-Method
tagger NN O I-Method
seems NN O O
to NN O O
be NN O O
superior NN O O
even NN O O
with NN O O
the NN O O
limited NN O O
time NN O O
we NN O O
allowed NN O O
ourselves NN O O
for NN O O
rule NN O O
development NN O O
. NN O O

-DOCSTART- NN O O

Structured-light NN O I-Method
methods NN O I-Method
actively NN O O
generate NN O O
geometric NN O I-Material
correspondence NN O I-Material
data NN O I-Material
between NN O O
projectors NN O O
and NN O O
cameras NN O O
in NN O O
order NN O O
to NN O O
facilitate NN O O
robust NN O I-Task
3D NN O I-Task
reconstruction NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
Photogeometric NN O I-Method
Structured NN O I-Method
Light NN O I-Method
whereby NN O O
a NN O O
standard NN O O
structured NN O I-Method
light NN O I-Method
method NN O I-Method
is NN O O
extended NN O O
to NN O O
include NN O O
photometric NN O I-Method
methods NN O I-Method
. NN O O

Photometric NN O I-Method
processing NN O I-Method
serves NN O O
the NN O O
double NN O O
purpose NN O O
of NN O O
increasing NN O O
the NN O O
amount NN O O
of NN O O
recovered NN O I-OtherScientificTerm
surface NN O I-OtherScientificTerm
detail NN O I-OtherScientificTerm
and NN O O
of NN O O
enabling NN O O
the NN O O
structured-light NN O I-OtherScientificTerm
setup NN O I-OtherScientificTerm
to NN O O
be NN O O
robustly NN O O
self-calibrated NN O O
. NN O O

Further NN O O
, NN O O
our NN O O
framework NN O I-Generic
uses NN O O
a NN O O
photogeometric NN O I-Method
optimization NN O I-Method
that NN O O
supports NN O O
the NN O O
simultaneous NN O O
use NN O O
of NN O O
multiple NN O O
cameras NN O O
and NN O O
projectors NN O O
and NN O O
yields NN O O
a NN O O
single NN O O
and NN O O
accurate NN O O
multi-view NN O I-Method
3D NN O I-Method
model NN O I-Method
which NN O O
best NN O O
complies NN O O
with NN O O
photometric NN O I-Material
and NN O I-Material
geometric NN O I-Material
data NN O I-Material
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
a NN O O
discrimination NN O O
and NN O O
robustness NN O O
oriented NN O O
adaptive NN O I-Method
learning NN O I-Method
procedure NN O I-Method
is NN O O
proposed NN O O
to NN O O
deal NN O O
with NN O O
the NN O O
task NN O O
of NN O O
syntactic NN O I-Task
ambiguity NN O I-Task
resolution NN O I-Task
. NN O O

Owing NN O O
to NN O O
the NN O O
problem NN O O
of NN O O
insufficient NN O I-OtherScientificTerm
training NN O I-OtherScientificTerm
data NN O I-OtherScientificTerm
and NN O O
approximation NN O I-OtherScientificTerm
error NN O I-OtherScientificTerm
introduced NN O O
by NN O O
the NN O O
language NN O I-Method
model NN O I-Method
, NN O O
traditional NN O O
statistical NN O I-Method
approaches NN O I-Method
, NN O O
which NN O O
resolve NN O O
ambiguities NN O I-OtherScientificTerm
by NN O O
indirectly NN O O
and NN O O
implicitly NN O O
using NN O O
maximum NN O I-Method
likelihood NN O I-Method
method NN O I-Method
, NN O O
fail NN O O
to NN O O
achieve NN O O
high NN O O
performance NN O O
in NN O O
real NN O O
applications NN O O
. NN O O

The NN O O
proposed NN O O
method NN O I-Generic
remedies NN O O
these NN O O
problems NN O I-Generic
by NN O O
adjusting NN O O
the NN O O
parameters NN O O
to NN O O
maximize NN O O
the NN O O
accuracy NN O O
rate NN O O
directly NN O O
. NN O O

To NN O O
make NN O O
the NN O O
proposed NN O O
algorithm NN O I-Generic
robust NN O O
, NN O O
the NN O O
possible NN O O
variations NN O O
between NN O O
the NN O O
training NN O O
corpus NN O O
and NN O O
the NN O O
real NN O O
tasks NN O O
are NN O O
also NN O O
taken NN O O
into NN O O
consideration NN O O
by NN O O
enlarging NN O O
the NN O O
separation NN O I-OtherScientificTerm
margin NN O I-OtherScientificTerm
between NN O O
the NN O O
correct NN O O
candidate NN O O
and NN O O
its NN O O
competing NN O O
members NN O O
. NN O O

Significant NN O O
improvement NN O O
has NN O O
been NN O O
observed NN O O
in NN O O
the NN O O
test NN O O
. NN O O

The NN O O
accuracy NN O I-Metric
rate NN O I-Metric
of NN O O
syntactic NN O I-Task
disambiguation NN O I-Task
is NN O O
raised NN O O
from NN O O
46.0 NN O O
% NN O O
to NN O O
60.62 NN O O
% NN O O
by NN O O
using NN O O
this NN O O
novel NN O O
approach NN O I-Generic
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
new NN O O
approach NN O I-Generic
to NN O O
statistical NN O I-Task
sentence NN O I-Task
generation NN O I-Task
in NN O O
which NN O O
alternative NN O O
phrases NN O O
are NN O O
represented NN O O
as NN O O
packed NN O O
sets NN O O
of NN O O
trees NN O I-OtherScientificTerm
, NN O O
or NN O O
forests NN O I-OtherScientificTerm
, NN O O
and NN O O
then NN O O
ranked NN O O
statistically NN O O
to NN O O
choose NN O O
the NN O O
best NN O O
one NN O O
. NN O O

This NN O O
representation NN O I-Generic
offers NN O O
advantages NN O O
in NN O O
compactness NN O O
and NN O O
in NN O O
the NN O O
ability NN O O
to NN O O
represent NN O O
syntactic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
. NN O O

It NN O I-Generic
also NN O O
facilitates NN O O
more NN O O
efficient NN O O
statistical NN O I-Method
ranking NN O I-Method
than NN O O
a NN O O
previous NN O O
approach NN O I-Generic
to NN O O
statistical NN O I-Task
generation NN O I-Task
. NN O O

An NN O O
efficient NN O O
ranking NN O I-Method
algorithm NN O I-Method
is NN O O
described NN O O
, NN O O
together NN O O
with NN O O
experimental NN O O
results NN O O
showing NN O O
significant NN O O
improvements NN O O
over NN O O
simple NN O O
enumeration NN O I-Method
or NN O O
a NN O O
lattice-based NN O I-Method
approach NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
article NN O O
deals NN O O
with NN O O
the NN O O
interpretation NN O O
of NN O O
conceptual NN O I-Task
operations NN O I-Task
underlying NN O O
the NN O O
communicative NN O O
use NN O O
of NN O O
natural NN O I-Material
language NN O I-Material
-LRB- NN O I-Material
NL NN O I-Material
-RRB- NN O I-Material
within NN O O
the NN O O
Structured NN O I-Method
Inheritance NN O I-Method
Network NN O I-Method
-LRB- NN O I-Method
SI-Nets NN O I-Method
-RRB- NN O I-Method
paradigm NN O I-Method
. NN O O

The NN O O
operations NN O I-Generic
are NN O O
reduced NN O O
to NN O O
functions NN O O
of NN O O
a NN O O
formal NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
, NN O O
thus NN O O
changing NN O O
the NN O O
level NN O O
of NN O O
abstraction NN O O
of NN O O
the NN O O
operations NN O I-Generic
to NN O O
be NN O O
performed NN O O
on NN O O
SI-Nets NN O I-Method
. NN O O

In NN O O
this NN O O
sense NN O O
, NN O O
operations NN O I-Generic
on NN O O
SI-Nets NN O I-Method
are NN O O
not NN O O
merely NN O O
isomorphic NN O O
to NN O O
single NN O O
epistemological NN O I-OtherScientificTerm
objects NN O I-OtherScientificTerm
, NN O O
but NN O O
can NN O O
be NN O O
viewed NN O O
as NN O O
a NN O O
simulation NN O O
of NN O O
processes NN O O
on NN O O
a NN O O
different NN O O
level NN O O
, NN O O
that NN O O
pertaining NN O O
to NN O O
the NN O O
conceptual NN O I-Method
system NN O I-Method
of NN O O
NL NN O I-Material
. NN O O

For NN O O
this NN O O
purpose NN O O
, NN O O
we NN O O
have NN O O
designed NN O O
a NN O O
version NN O O
of NN O O
KL-ONE NN O I-Method
which NN O O
represents NN O O
the NN O O
epistemological NN O I-OtherScientificTerm
level NN O I-OtherScientificTerm
, NN O O
while NN O O
the NN O O
new NN O O
experimental NN O I-Generic
language NN O I-Generic
, NN O O
KL-Conc NN O I-Method
, NN O O
represents NN O O
the NN O O
conceptual NN O I-OtherScientificTerm
level NN O I-OtherScientificTerm
. NN O O

KL-Conc NN O I-Method
would NN O O
seem NN O O
to NN O O
be NN O O
a NN O O
more NN O O
natural NN O O
and NN O O
intuitive NN O O
way NN O O
of NN O O
interacting NN O O
with NN O O
SI-Nets NN O I-Method
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
an NN O O
algorithm NN O I-Generic
for NN O O
calibrated NN O I-Task
camera NN O I-Task
relative NN O I-Task
pose NN O I-Task
estimation NN O I-Task
from NN O O
lines NN O O
. NN O O

Given NN O O
three NN O O
lines NN O O
with NN O O
two NN O O
of NN O O
the NN O O
lines NN O O
parallel NN O O
and NN O O
orthogonal NN O O
to NN O O
the NN O O
third NN O O
we NN O O
can NN O O
compute NN O O
the NN O O
relative NN O I-OtherScientificTerm
rotation NN O I-OtherScientificTerm
between NN O O
two NN O O
images NN O O
. NN O O

We NN O O
can NN O O
also NN O O
compute NN O O
the NN O O
relative NN O I-OtherScientificTerm
translation NN O I-OtherScientificTerm
from NN O O
two NN O O
intersection NN O O
points NN O O
. NN O O

We NN O O
also NN O O
present NN O O
a NN O O
framework NN O I-Generic
in NN O O
which NN O O
such NN O O
lines NN O O
can NN O O
be NN O O
detected NN O O
. NN O O

We NN O O
evaluate NN O O
the NN O O
performance NN O O
of NN O O
the NN O O
algorithm NN O I-Generic
using NN O O
synthetic NN O I-Material
and NN O I-Material
real NN O I-Material
data NN O I-Material
. NN O O

The NN O O
intended NN O O
use NN O O
of NN O O
the NN O O
algorithm NN O I-Generic
is NN O O
with NN O O
robust NN O O
hypothesize-and-test NN O I-Method
frameworks NN O I-Method
such NN O O
as NN O O
RANSAC NN O I-Method
. NN O O

Our NN O O
approach NN O I-Generic
is NN O O
suitable NN O O
for NN O O
urban NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
indoor NN O I-OtherScientificTerm
environments NN O I-OtherScientificTerm
where NN O O
most NN O O
lines NN O O
are NN O O
either NN O O
parallel NN O O
or NN O O
orthogonal NN O O
to NN O O
each NN O O
other NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
a NN O O
fully NN O I-Method
automated NN O I-Method
extraction NN O I-Method
system NN O I-Method
, NN O O
named NN O O
IntEx NN O I-Method
, NN O O
to NN O O
identify NN O O
gene NN O I-Task
and NN O I-Task
protein NN O I-Task
interactions NN O I-Task
in NN O O
biomedical NN O I-Material
text NN O I-Material
. NN O O

Our NN O O
approach NN O I-Generic
is NN O O
based NN O O
on NN O O
first NN O O
splitting NN O O
complex NN O O
sentences NN O O
into NN O O
simple NN O O
clausal NN O O
structures NN O O
made NN O O
up NN O O
of NN O O
syntactic NN O I-OtherScientificTerm
roles NN O I-OtherScientificTerm
. NN O O

Then NN O O
, NN O O
tagging NN O O
biological NN O I-OtherScientificTerm
entities NN O I-OtherScientificTerm
with NN O O
the NN O O
help NN O O
of NN O O
biomedical NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
linguistic NN O I-OtherScientificTerm
ontologies NN O I-OtherScientificTerm
. NN O O

Finally NN O O
, NN O O
extracting NN O O
complete NN O O
interactions NN O O
by NN O O
analyzing NN O O
the NN O O
matching NN O O
contents NN O O
of NN O O
syntactic NN O I-OtherScientificTerm
roles NN O I-OtherScientificTerm
and NN O O
their NN O O
linguistically NN O O
significant NN O O
combinations NN O O
. NN O O

Our NN O O
extraction NN O I-Method
system NN O I-Method
handles NN O O
complex NN O O
sentences NN O O
and NN O O
extracts NN O O
multiple NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
nested NN O I-OtherScientificTerm
interactions NN O I-OtherScientificTerm
specified NN O O
in NN O O
a NN O O
sentence NN O O
. NN O O

Experimental NN O O
evaluations NN O O
with NN O O
two NN O O
other NN O O
state NN O O
of NN O O
the NN O O
art NN O O
extraction NN O I-Method
systems NN O I-Method
indicate NN O O
that NN O O
the NN O O
IntEx NN O I-Method
system NN O I-Method
achieves NN O O
better NN O O
performance NN O O
without NN O O
the NN O O
labor NN O O
intensive NN O O
pattern NN O I-OtherScientificTerm
engineering NN O I-OtherScientificTerm
requirement NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
introduces NN O O
a NN O O
method NN O I-Generic
for NN O O
computational NN O I-Task
analysis NN O I-Task
of NN O I-Task
move NN O I-Task
structures NN O I-Task
in NN O O
abstracts NN O I-Material
of NN O I-Material
research NN O I-Material
articles NN O I-Material
. NN O O

In NN O O
our NN O O
approach NN O I-Generic
, NN O O
sentences NN O O
in NN O O
a NN O O
given NN O O
abstract NN O O
are NN O O
analyzed NN O O
and NN O O
labeled NN O O
with NN O O
a NN O O
specific NN O O
move NN O O
in NN O O
light NN O O
of NN O O
various NN O O
rhetorical NN O I-OtherScientificTerm
functions NN O I-OtherScientificTerm
. NN O O

The NN O O
method NN O I-Generic
involves NN O O
automatically NN O O
gathering NN O O
a NN O O
large NN O O
number NN O O
of NN O O
abstracts NN O I-Material
from NN O O
the NN O O
Web NN O I-Material
and NN O O
building NN O O
a NN O O
language NN O I-Method
model NN O I-Method
of NN O O
abstract NN O I-Material
moves NN O I-Material
. NN O O

We NN O O
also NN O O
present NN O O
a NN O O
prototype NN O I-Method
concordancer NN O I-Method
, NN O O
CARE NN O I-Method
, NN O O
which NN O O
exploits NN O O
the NN O O
move-tagged NN O I-Material
abstracts NN O I-Material
for NN O O
digital NN O I-Task
learning NN O I-Task
. NN O O

This NN O O
system NN O I-Generic
provides NN O O
a NN O O
promising NN O O
approach NN O I-Generic
to NN O O
Web-based NN O I-Task
computer-assisted NN O I-Task
academic NN O I-Task
writing NN O I-Task
. NN O O

-DOCSTART- NN O O

This NN O O
work NN O O
presents NN O O
a NN O O
real-time NN O I-Method
system NN O I-Method
for NN O O
multiple NN O I-Task
object NN O I-Task
tracking NN O I-Task
in NN O I-Task
dynamic NN O I-Task
scenes NN O I-Task
. NN O O

A NN O O
unique NN O O
characteristic NN O O
of NN O O
the NN O O
system NN O I-Generic
is NN O O
its NN O O
ability NN O O
to NN O O
cope NN O O
with NN O O
long-duration NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
complete NN O I-OtherScientificTerm
occlusion NN O I-OtherScientificTerm
without NN O O
a NN O O
prior NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
about NN O O
the NN O O
shape NN O I-OtherScientificTerm
or NN O O
motion NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
objects NN O I-OtherScientificTerm
. NN O O

The NN O O
system NN O I-Generic
produces NN O O
good NN O O
segment NN O O
and NN O O
tracking NN O I-Task
results NN O O
at NN O O
a NN O O
frame NN O I-OtherScientificTerm
rate NN O I-OtherScientificTerm
of NN O O
15-20 NN O O
fps NN O O
for NN O O
image NN O I-OtherScientificTerm
size NN O I-OtherScientificTerm
of NN O O
320x240 NN O O
, NN O O
as NN O O
demonstrated NN O O
by NN O O
extensive NN O O
experiments NN O O
performed NN O O
using NN O O
video NN O I-Material
sequences NN O I-Material
under NN O O
different NN O O
conditions NN O O
indoor NN O O
and NN O O
outdoor NN O O
with NN O O
long-duration NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
complete NN O I-OtherScientificTerm
occlusions NN O I-OtherScientificTerm
in NN O O
changing NN O I-OtherScientificTerm
background NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
a NN O O
method NN O I-Generic
of NN O O
organizing NN O I-Task
reading NN O I-Task
materials NN O I-Task
for NN O O
vocabulary NN O I-Task
learning NN O I-Task
. NN O O

It NN O I-Generic
enables NN O O
us NN O O
to NN O O
select NN O O
a NN O O
concise NN O O
set NN O O
of NN O O
reading NN O O
texts NN O O
-LRB- NN O O
from NN O O
a NN O O
target NN O O
corpus NN O O
-RRB- NN O O
that NN O O
contains NN O O
all NN O O
the NN O O
target NN O I-OtherScientificTerm
vocabulary NN O I-OtherScientificTerm
to NN O O
be NN O O
learned NN O O
. NN O O

We NN O O
used NN O O
a NN O O
specialized NN O O
vocabulary NN O O
for NN O O
an NN O O
English NN O O
certification NN O O
test NN O O
as NN O O
the NN O O
target NN O O
vocabulary NN O O
and NN O O
used NN O O
English NN O I-Material
Wikipedia NN O I-Material
, NN O O
a NN O O
free-content NN O I-Material
encyclopedia NN O I-Material
, NN O O
as NN O O
the NN O O
target NN O O
corpus NN O O
. NN O O

The NN O O
organized NN O O
reading NN O O
materials NN O O
would NN O O
enable NN O O
learners NN O O
not NN O O
only NN O O
to NN O O
study NN O O
the NN O O
target NN O O
vocabulary NN O O
efficiently NN O O
but NN O O
also NN O O
to NN O O
gain NN O O
a NN O O
variety NN O O
of NN O O
knowledge NN O O
through NN O O
reading NN O O
. NN O O

The NN O O
reading NN O O
materials NN O O
are NN O O
available NN O O
on NN O O
our NN O O
web NN O O
site NN O O
. NN O O

-DOCSTART- NN O O

A NN O O
novel NN O O
bootstrapping NN O I-Method
approach NN O I-Method
to NN O O
Named NN O I-Task
Entity NN O I-Task
-LRB- NN O I-Task
NE NN O I-Task
-RRB- NN O I-Task
tagging NN O I-Task
using NN O O
concept-based NN O I-OtherScientificTerm
seeds NN O I-OtherScientificTerm
and NN O O
successive NN O I-Method
learners NN O I-Method
is NN O O
presented NN O O
. NN O O

This NN O O
approach NN O I-Generic
only NN O O
requires NN O O
a NN O O
few NN O O
common NN O O
noun NN O O
or NN O O
pronoun NN O O
seeds NN O O
that NN O O
correspond NN O O
to NN O O
the NN O O
concept NN O O
for NN O O
the NN O O
targeted NN O O
NE NN O I-OtherScientificTerm
, NN O O
e.g. NN O O
he/she/man NN O O
/ NN O O
woman NN O O
for NN O O
PERSON NN O I-OtherScientificTerm
NE NN O I-OtherScientificTerm
. NN O O

The NN O O
bootstrapping NN O I-Method
procedure NN O I-Method
is NN O O
implemented NN O O
as NN O O
training NN O O
two NN O O
successive NN O I-Method
learners NN O I-Method
. NN O O

First NN O O
, NN O O
decision NN O I-OtherScientificTerm
list NN O I-OtherScientificTerm
is NN O O
used NN O O
to NN O O
learn NN O O
the NN O O
parsing-based NN O I-OtherScientificTerm
NE NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
. NN O O

Then NN O O
, NN O O
a NN O O
Hidden NN O I-Method
Markov NN O I-Method
Model NN O I-Method
is NN O O
trained NN O O
on NN O O
a NN O O
corpus NN O O
automatically NN O O
tagged NN O O
by NN O O
the NN O O
first NN O O
learner NN O I-Generic
. NN O O

The NN O O
resulting NN O O
NE NN O I-Method
system NN O I-Method
approaches NN O O
supervised NN O I-Task
NE NN O I-Task
performance NN O O
for NN O O
some NN O O
NE NN O O
types NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
the NN O O
first NN O O
known NN O O
empirical NN O O
test NN O O
of NN O O
an NN O O
increasingly NN O O
common NN O O
speculative NN O O
claim NN O O
, NN O O
by NN O O
evaluating NN O O
a NN O O
representative NN O O
Chinese-to-English NN O I-Method
SMT NN O I-Method
model NN O I-Method
directly NN O O
on NN O O
word NN O I-Task
sense NN O I-Task
disambiguation NN O I-Task
performance NN O O
, NN O O
using NN O O
standard NN O O
WSD NN O I-Method
evaluation NN O I-Method
methodology NN O I-Method
and NN O O
datasets NN O O
from NN O O
the NN O O
Senseval-3 NN O I-Material
Chinese NN O I-Material
lexical NN O I-Material
sample NN O I-Material
task NN O I-Material
. NN O O

Much NN O O
effort NN O O
has NN O O
been NN O O
put NN O O
in NN O O
designing NN O O
and NN O O
evaluating NN O O
dedicated NN O I-Method
word NN O I-Method
sense NN O I-Method
disambiguation NN O I-Method
-LRB- NN O I-Method
WSD NN O I-Method
-RRB- NN O I-Method
models NN O I-Method
, NN O O
in NN O O
particular NN O O
with NN O O
the NN O O
Senseval NN O I-Material
series NN O I-Material
of NN O I-Material
workshops NN O I-Material
. NN O O

At NN O O
the NN O O
same NN O O
time NN O O
, NN O O
the NN O O
recent NN O O
improvements NN O O
in NN O O
the NN O O
BLEU NN O I-Metric
scores NN O I-Metric
of NN O O
statistical NN O I-Task
machine NN O I-Task
translation NN O I-Task
-LRB- NN O I-Task
SMT NN O I-Task
-RRB- NN O I-Task
suggests NN O O
that NN O O
SMT NN O I-Method
models NN O I-Method
are NN O O
good NN O O
at NN O O
predicting NN O O
the NN O O
right NN O O
translation NN O I-Task
of NN O O
the NN O O
words NN O O
in NN O O
source NN O O
language NN O O
sentences NN O O
. NN O O

Surprisingly NN O O
however NN O O
, NN O O
the NN O O
WSD NN O I-Metric
accuracy NN O I-Metric
of NN O O
SMT NN O I-Method
models NN O I-Method
has NN O O
never NN O O
been NN O O
evaluated NN O O
and NN O O
compared NN O O
with NN O O
that NN O I-Generic
of NN O O
the NN O O
dedicated NN O I-Method
WSD NN O I-Method
models NN O I-Method
. NN O O

We NN O O
present NN O O
controlled NN O O
experiments NN O O
showing NN O O
the NN O O
WSD NN O I-Metric
accuracy NN O I-Metric
of NN O O
current NN O O
typical NN O O
SMT NN O I-Method
models NN O I-Method
to NN O O
be NN O O
significantly NN O O
lower NN O O
than NN O O
that NN O I-Generic
of NN O O
all NN O O
the NN O O
dedicated NN O I-Method
WSD NN O I-Method
models NN O I-Method
considered NN O O
. NN O O

This NN O O
tends NN O O
to NN O O
support NN O O
the NN O O
view NN O O
that NN O O
despite NN O O
recent NN O O
speculative NN O O
claims NN O O
to NN O O
the NN O O
contrary NN O O
, NN O O
current NN O O
SMT NN O I-Method
models NN O I-Method
do NN O O
have NN O O
limitations NN O O
in NN O O
comparison NN O O
with NN O O
dedicated NN O I-Method
WSD NN O I-Method
models NN O I-Method
, NN O O
and NN O O
that NN O O
SMT NN O I-Method
should NN O O
benefit NN O O
from NN O O
the NN O O
better NN O O
predictions NN O O
made NN O O
by NN O O
the NN O O
WSD NN O I-Method
models NN O I-Method
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
present NN O O
a NN O O
novel NN O O
, NN O O
customizable NN O O
: NN O O
IE NN O I-Method
paradigm NN O I-Method
that NN O O
takes NN O O
advantage NN O O
of NN O O
predicate-argument NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
. NN O O

We NN O O
also NN O O
introduce NN O O
a NN O O
new NN O O
way NN O O
of NN O O
automatically NN O I-Task
identifying NN O I-Task
predicate NN O I-Task
argument NN O I-Task
structures NN O I-Task
, NN O O
which NN O O
is NN O O
central NN O O
to NN O O
our NN O O
IE NN O I-Method
paradigm NN O I-Method
. NN O O

It NN O I-Generic
is NN O O
based NN O O
on NN O O
: NN O O
-LRB- NN O O
1 NN O O
-RRB- NN O O
an NN O O
extended NN O O
set NN O O
of NN O O
features NN O I-OtherScientificTerm
; NN O O
and NN O O
-LRB- NN O O
2 NN O O
-RRB- NN O O
inductive NN O I-Method
decision NN O I-Method
tree NN O I-Method
learning NN O I-Method
. NN O O

The NN O O
experimental NN O O
results NN O O
prove NN O O
our NN O O
claim NN O O
that NN O O
accurate NN O O
predicate-argument NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
enable NN O O
high NN O O
quality NN O O
IE NN O I-Task
results NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
present NN O O
a NN O O
statistical NN O I-Generic
profile NN O I-Generic
of NN O O
the NN O O
Named NN O I-Task
Entity NN O I-Task
task NN O I-Task
, NN O O
a NN O O
specific NN O O
information NN O I-Task
extraction NN O I-Task
task NN O I-Task
for NN O O
which NN O O
corpora NN O O
in NN O O
several NN O O
languages NN O O
are NN O O
available NN O O
. NN O O

Using NN O O
the NN O O
results NN O O
of NN O O
the NN O O
statistical NN O I-Method
analysis NN O I-Method
, NN O O
we NN O O
propose NN O O
an NN O O
algorithm NN O I-Generic
for NN O O
lower NN O I-Method
bound NN O I-Method
estimation NN O I-Method
for NN O O
Named NN O I-Material
Entity NN O I-Material
corpora NN O I-Material
and NN O O
discuss NN O O
the NN O O
significance NN O O
of NN O O
the NN O O
cross-lingual NN O O
comparisons NN O O
provided NN O O
by NN O O
the NN O O
analysis NN O I-Generic
. NN O O

-DOCSTART- NN O O

We NN O O
attack NN O O
an NN O O
inexplicably NN O O
under-explored NN O I-Material
language NN O I-Material
genre NN O I-Material
of NN O I-Material
spoken NN O I-Material
language NN O I-Material
-- NN O O
lyrics NN O I-Material
in NN O I-Material
music NN O I-Material
-- NN O O
via NN O O
completely NN O O
unsuper-vised NN O I-Method
induction NN O I-Method
of NN O O
an NN O O
SMT-style NN O I-Method
stochastic NN O I-Method
transduction NN O I-Method
grammar NN O I-Method
for NN O O
hip NN O I-Material
hop NN O I-Material
lyrics NN O I-Material
, NN O O
yielding NN O O
a NN O O
fully-automatically NN O I-Method
learned NN O I-Method
challenge-response NN O I-Method
system NN O I-Method
that NN O O
produces NN O O
rhyming NN O I-Material
lyrics NN O I-Material
given NN O O
an NN O O
input NN O O
. NN O O

Unlike NN O O
previous NN O O
efforts NN O O
, NN O O
we NN O O
choose NN O O
the NN O O
domain NN O O
of NN O O
hip NN O I-Material
hop NN O I-Material
lyrics NN O I-Material
, NN O O
which NN O O
is NN O O
particularly NN O O
unstructured NN O O
and NN O O
noisy NN O O
. NN O O

A NN O O
novel NN O O
feature NN O O
of NN O O
our NN O O
approach NN O I-Generic
is NN O O
that NN O O
it NN O O
is NN O O
completely NN O O
unsupervised NN O O
and NN O O
requires NN O O
no NN O O
a NN O O
priori NN O I-OtherScientificTerm
linguistic NN O I-OtherScientificTerm
or NN O I-OtherScientificTerm
phonetic NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
. NN O O

In NN O O
spite NN O O
of NN O O
the NN O O
level NN O O
of NN O O
difficulty NN O O
of NN O O
the NN O O
challenge NN O O
, NN O O
the NN O O
model NN O I-Generic
nevertheless NN O O
produces NN O O
fluent NN O O
output NN O O
as NN O O
judged NN O O
by NN O O
human NN O I-OtherScientificTerm
evaluators NN O I-OtherScientificTerm
, NN O O
and NN O O
performs NN O O
significantly NN O O
better NN O O
than NN O O
widely NN O O
used NN O O
phrase-based NN O I-Method
SMT NN O I-Method
models NN O I-Method
upon NN O O
the NN O O
same NN O O
task NN O I-Generic
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
investigate NN O O
the NN O O
problem NN O O
of NN O O
automatically NN O O
predicting NN O I-Task
segment NN O I-Task
boundaries NN O I-Task
in NN O O
spoken NN O I-Material
multiparty NN O I-Material
dialogue NN O I-Material
. NN O O

We NN O O
extend NN O O
prior NN O O
work NN O O
in NN O O
two NN O O
ways NN O O
. NN O O

We NN O O
first NN O O
apply NN O O
approaches NN O I-Generic
that NN O O
have NN O O
been NN O O
proposed NN O O
for NN O O
predicting NN O I-Task
top-level NN O I-Task
topic NN O I-Task
shifts NN O I-Task
to NN O O
the NN O O
problem NN O O
of NN O O
identifying NN O I-Task
subtopic NN O I-Task
boundaries NN O I-Task
. NN O O

We NN O O
then NN O O
explore NN O O
the NN O O
impact NN O O
on NN O O
performance NN O O
of NN O O
using NN O O
ASR NN O I-OtherScientificTerm
output NN O I-OtherScientificTerm
as NN O O
opposed NN O O
to NN O O
human NN O I-OtherScientificTerm
transcription NN O I-OtherScientificTerm
. NN O O

Examination NN O O
of NN O O
the NN O O
effect NN O O
of NN O O
features NN O I-OtherScientificTerm
shows NN O O
that NN O O
predicting NN O I-Task
top-level NN O I-Task
and NN O I-Task
predicting NN O I-Task
subtopic NN O I-Task
boundaries NN O I-Task
are NN O O
two NN O O
distinct NN O O
tasks NN O I-Generic
: NN O O
-LRB- NN O O
1 NN O O
-RRB- NN O O
for NN O O
predicting NN O I-Task
subtopic NN O I-Task
boundaries NN O I-Task
, NN O O
the NN O O
lexical NN O I-Method
cohesion-based NN O I-Method
approach NN O I-Method
alone NN O O
can NN O O
achieve NN O O
competitive NN O O
results NN O O
, NN O O
-LRB- NN O O
2 NN O O
-RRB- NN O O
for NN O O
predicting NN O I-Task
top-level NN O I-Task
boundaries NN O I-Task
, NN O O
the NN O O
machine NN O I-Method
learning NN O I-Method
approach NN O I-Method
that NN O O
combines NN O O
lexical-cohesion NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
conversational NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
performs NN O O
best NN O O
, NN O O
and NN O O
-LRB- NN O O
3 NN O O
-RRB- NN O O
conversational NN O I-OtherScientificTerm
cues NN O I-OtherScientificTerm
, NN O O
such NN O O
as NN O O
cue NN O I-OtherScientificTerm
phrases NN O I-OtherScientificTerm
and NN O O
overlapping NN O I-OtherScientificTerm
speech NN O I-OtherScientificTerm
, NN O O
are NN O O
better NN O O
indicators NN O I-Generic
for NN O O
the NN O O
top-level NN O I-Generic
prediction NN O I-Generic
task NN O I-Generic
. NN O O

We NN O O
also NN O O
find NN O O
that NN O O
the NN O O
transcription NN O I-OtherScientificTerm
errors NN O I-OtherScientificTerm
inevitable NN O O
in NN O O
ASR NN O I-OtherScientificTerm
output NN O I-OtherScientificTerm
have NN O O
a NN O O
negative NN O O
impact NN O O
on NN O O
models NN O I-Generic
that NN O O
combine NN O O
lexical-cohesion NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
conversational NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
, NN O O
but NN O O
do NN O O
not NN O O
change NN O O
the NN O O
general NN O O
preference NN O O
of NN O O
approach NN O O
for NN O O
the NN O O
two NN O O
tasks NN O I-Generic
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
a NN O O
simple NN O O
unsupervised NN O I-Method
technique NN O I-Method
for NN O O
learning NN O O
morphology NN O I-Task
by NN O O
identifying NN O O
hubs NN O I-OtherScientificTerm
in NN O O
an NN O O
automaton NN O I-OtherScientificTerm
. NN O O

For NN O O
our NN O O
purposes NN O O
, NN O O
a NN O O
hub NN O I-OtherScientificTerm
is NN O O
a NN O O
node NN O I-OtherScientificTerm
in NN O O
a NN O O
graph NN O I-OtherScientificTerm
with NN O O
in-degree NN O O
greater NN O O
than NN O O
one NN O O
and NN O O
out-degree NN O O
greater NN O O
than NN O O
one NN O O
. NN O O

We NN O O
create NN O O
a NN O O
word-trie NN O I-Method
, NN O O
transform NN O O
it NN O I-Generic
into NN O O
a NN O O
minimal NN O I-OtherScientificTerm
DFA NN O I-OtherScientificTerm
, NN O O
then NN O O
identify NN O O
hubs NN O I-OtherScientificTerm
. NN O O

Those NN O O
hubs NN O I-OtherScientificTerm
mark NN O O
the NN O O
boundary NN O O
between NN O O
root NN O I-OtherScientificTerm
and NN O O
suffix NN O I-OtherScientificTerm
, NN O O
achieving NN O O
similar NN O O
performance NN O O
to NN O O
more NN O O
complex NN O O
mixtures NN O O
of NN O O
techniques NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
Bayesian NN O I-Task
machine NN O I-Task
learning NN O I-Task
, NN O O
conjugate NN O I-OtherScientificTerm
priors NN O I-OtherScientificTerm
are NN O O
popular NN O O
, NN O O
mostly NN O O
due NN O O
to NN O O
mathematical NN O O
convenience NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
show NN O O
that NN O O
there NN O O
are NN O O
deeper NN O O
reasons NN O O
for NN O O
choosing NN O O
a NN O O
conjugate NN O I-OtherScientificTerm
prior NN O I-OtherScientificTerm
. NN O O

Specifically NN O O
, NN O O
we NN O O
formulate NN O O
the NN O O
conjugate NN O I-OtherScientificTerm
prior NN O I-OtherScientificTerm
in NN O O
the NN O O
form NN O O
of NN O O
Bregman NN O I-OtherScientificTerm
divergence NN O I-OtherScientificTerm
and NN O O
show NN O O
that NN O O
it NN O O
is NN O O
the NN O O
inherent NN O O
geometry NN O O
of NN O O
conjugate NN O I-OtherScientificTerm
priors NN O I-OtherScientificTerm
that NN O O
makes NN O O
them NN O O
appropriate NN O O
and NN O O
intuitive NN O O
. NN O O

This NN O O
geometric NN O O
interpretation NN O O
allows NN O O
one NN O O
to NN O O
view NN O O
the NN O O
hyperparameters NN O O
of NN O O
conjugate NN O I-OtherScientificTerm
priors NN O I-OtherScientificTerm
as NN O O
the NN O O
effective NN O O
sample NN O O
points NN O O
, NN O O
thus NN O O
providing NN O O
additional NN O O
intuition NN O O
. NN O O

We NN O O
use NN O O
this NN O O
geometric NN O I-OtherScientificTerm
understanding NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
conjugate NN O I-OtherScientificTerm
priors NN O I-OtherScientificTerm
to NN O O
derive NN O O
the NN O O
hyperparameters NN O I-Generic
and NN O O
expression NN O O
of NN O O
the NN O O
prior NN O I-Generic
used NN O O
to NN O O
couple NN O O
the NN O O
generative NN O I-Method
and NN O I-Method
discriminative NN O I-Method
components NN O I-Method
of NN O O
a NN O O
hybrid NN O I-Method
model NN O I-Method
for NN O O
semi-supervised NN O I-Task
learning NN O I-Task
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
defines NN O O
a NN O O
generative NN O I-Method
probabilistic NN O I-Method
model NN O I-Method
of NN O I-Method
parse NN O I-Method
trees NN O I-Method
, NN O O
which NN O O
we NN O O
call NN O O
PCFG-LA NN O I-Method
. NN O O

This NN O O
model NN O I-Generic
is NN O O
an NN O O
extension NN O O
of NN O O
PCFG NN O I-Method
in NN O O
which NN O O
non-terminal NN O I-OtherScientificTerm
symbols NN O I-OtherScientificTerm
are NN O O
augmented NN O O
with NN O O
latent NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
. NN O O

Finegrained NN O I-OtherScientificTerm
CFG NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
are NN O O
automatically NN O O
induced NN O O
from NN O O
a NN O O
parsed NN O I-Material
corpus NN O I-Material
by NN O O
training NN O O
a NN O O
PCFG-LA NN O I-Method
model NN O I-Method
using NN O O
an NN O O
EM-algorithm NN O I-Method
. NN O O

Because NN O O
exact NN O I-Task
parsing NN O I-Task
with NN O O
a NN O O
PCFG-LA NN O I-Method
is NN O O
NP-hard NN O O
, NN O O
several NN O O
approximations NN O O
are NN O O
described NN O O
and NN O O
empirically NN O O
compared NN O O
. NN O O

In NN O O
experiments NN O O
using NN O O
the NN O O
Penn NN O I-Material
WSJ NN O I-Material
corpus NN O I-Material
, NN O O
our NN O O
automatically NN O O
trained NN O O
model NN O I-Generic
gave NN O O
a NN O O
performance NN O O
of NN O O
86.6 NN O O
% NN O O
-LRB- NN O O
F1 NN O I-Metric
, NN O O
sentences NN O O
< NN O O
40 NN O O
words NN O O
-RRB- NN O O
, NN O O
which NN O O
is NN O O
comparable NN O O
to NN O O
that NN O O
of NN O O
an NN O O
unlexicalized NN O I-Method
PCFG NN O I-Method
parser NN O I-Method
created NN O O
using NN O O
extensive NN O O
manual NN O I-Method
feature NN O I-Method
selection NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
reports NN O O
on NN O O
two NN O O
contributions NN O O
to NN O O
large NN O I-Task
vocabulary NN O I-Task
continuous NN O I-Task
speech NN O I-Task
recognition NN O I-Task
. NN O O

First NN O O
, NN O O
we NN O O
present NN O O
a NN O O
new NN O O
paradigm NN O O
for NN O O
speaker-independent NN O I-Task
-LRB- NN O I-Task
SI NN O I-Task
-RRB- NN O I-Task
training NN O I-Task
of NN O I-Task
hidden NN O I-Task
Markov NN O I-Task
models NN O I-Task
-LRB- NN O I-Task
HMM NN O I-Task
-RRB- NN O I-Task
, NN O O
which NN O O
uses NN O O
a NN O O
large NN O O
amount NN O O
of NN O O
speech NN O I-Material
from NN O O
a NN O O
few NN O O
speakers NN O O
instead NN O O
of NN O O
the NN O O
traditional NN O O
practice NN O O
of NN O O
using NN O O
a NN O O
little NN O O
speech NN O O
from NN O O
many NN O O
speakers NN O O
. NN O O

In NN O O
addition NN O O
, NN O O
combination NN O O
of NN O O
the NN O O
training NN O O
speakers NN O O
is NN O O
done NN O O
by NN O O
averaging NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
statistics NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
independently NN O I-OtherScientificTerm
trained NN O I-OtherScientificTerm
models NN O I-OtherScientificTerm
rather NN O O
than NN O O
the NN O O
usual NN O O
pooling NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
all NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
speech NN O I-OtherScientificTerm
data NN O I-OtherScientificTerm
from NN O O
many NN O O
speakers NN O O
prior NN O O
to NN O O
training NN O O
. NN O O

With NN O O
only NN O O
12 NN O O
training NN O O
speakers NN O O
for NN O O
SI NN O I-Task
recognition NN O I-Task
, NN O O
we NN O O
achieved NN O O
a NN O O
7.5 NN O O
% NN O O
word NN O I-Metric
error NN O I-Metric
rate NN O I-Metric
on NN O O
a NN O O
standard NN O O
grammar NN O O
and NN O O
test NN O O
set NN O O
from NN O O
the NN O O
DARPA NN O I-Material
Resource NN O I-Material
Management NN O I-Material
corpus NN O I-Material
. NN O O

This NN O O
performance NN O O
is NN O O
comparable NN O O
to NN O O
our NN O O
best NN O O
condition NN O O
for NN O O
this NN O O
test NN O O
suite NN O O
, NN O O
using NN O O
109 NN O O
training NN O O
speakers NN O O
. NN O O

Second NN O O
, NN O O
we NN O O
show NN O O
a NN O O
significant NN O O
improvement NN O O
for NN O O
speaker NN O I-Task
adaptation NN O I-Task
-LRB- NN O I-Task
SA NN O I-Task
-RRB- NN O I-Task
using NN O O
the NN O O
new NN O O
SI NN O I-Material
corpus NN O I-Material
and NN O O
a NN O O
small NN O O
amount NN O O
of NN O O
speech NN O O
from NN O O
the NN O O
new NN O O
-LRB- NN O O
target NN O O
-RRB- NN O O
speaker NN O O
. NN O O

A NN O O
probabilistic NN O I-Method
spectral NN O I-Method
mapping NN O I-Method
is NN O O
estimated NN O O
independently NN O O
for NN O O
each NN O O
training NN O O
-LRB- NN O O
reference NN O O
-RRB- NN O O
speaker NN O O
and NN O O
the NN O O
target NN O O
speaker NN O O
. NN O O

Each NN O O
reference NN O I-Method
model NN O I-Method
is NN O O
transformed NN O O
to NN O O
the NN O O
space NN O O
of NN O O
the NN O O
target NN O O
speaker NN O O
and NN O O
combined NN O O
by NN O O
averaging NN O O
. NN O O

Using NN O O
only NN O O
40 NN O O
utterances NN O O
from NN O O
the NN O O
target NN O O
speaker NN O O
for NN O O
adaptation NN O I-Task
, NN O O
the NN O O
error NN O I-Metric
rate NN O I-Metric
dropped NN O O
to NN O O
4.1 NN O O
% NN O O
-- NN O O
a NN O O
45 NN O O
% NN O O
reduction NN O O
in NN O O
error NN O O
compared NN O O
to NN O O
the NN O O
SI NN O I-Task
result NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
show NN O O
how NN O O
to NN O O
construct NN O O
a NN O O
transfer NN O I-Material
dictionary NN O I-Material
automatically NN O O
. NN O O

Dictionary NN O I-Task
construction NN O I-Task
, NN O O
one NN O O
of NN O O
the NN O O
most NN O O
difficult NN O O
tasks NN O O
in NN O O
developing NN O O
a NN O O
machine NN O I-Task
translation NN O I-Task
system NN O I-Task
, NN O O
is NN O O
expensive NN O O
. NN O O

To NN O O
avoid NN O O
this NN O O
problem NN O O
, NN O O
we NN O O
investigate NN O O
how NN O O
we NN O O
build NN O O
a NN O O
dictionary NN O I-Generic
using NN O O
existing NN O O
linguistic NN O I-Material
resources NN O I-Material
. NN O O

Our NN O O
algorithm NN O O
can NN O O
be NN O O
applied NN O O
to NN O O
any NN O O
language NN O O
pairs NN O O
, NN O O
but NN O O
for NN O O
the NN O O
present NN O O
we NN O O
focus NN O O
on NN O O
building NN O O
a NN O O
Korean-to-Japanese NN O I-Material
dictionary NN O I-Material
using NN O O
English NN O I-Material
as NN O O
a NN O O
pivot NN O O
. NN O O

We NN O O
attempt NN O O
three NN O O
ways NN O O
of NN O O
automatic NN O I-Task
construction NN O I-Task
to NN O O
corroborate NN O O
the NN O O
effect NN O O
of NN O O
the NN O O
directionality NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
dictionaries NN O I-Generic
. NN O O

First NN O O
, NN O O
we NN O O
introduce NN O O
`` NN O I-Method
one-time NN O I-Method
look NN O I-Method
up NN O I-Method
'' NN O I-Method
method NN O I-Method
using NN O O
a NN O O
Korean-to-English NN O I-Material
and NN O I-Material
a NN O I-Material
Japanese-to-English NN O I-Material
dictionary NN O I-Material
. NN O O

Second NN O O
, NN O O
we NN O O
show NN O O
a NN O O
method NN O I-Generic
using NN O O
`` NN O I-OtherScientificTerm
overlapping NN O I-OtherScientificTerm
constraint NN O I-OtherScientificTerm
'' NN O I-OtherScientificTerm
with NN O O
a NN O O
Korean-to-English NN O I-Material
dictionary NN O I-Material
and NN O O
an NN O O
English-to-Japanese NN O I-Material
dictionary NN O I-Material
. NN O O

Third NN O O
, NN O O
we NN O O
consider NN O O
another NN O O
alternative NN O O
method NN O I-Generic
rarely NN O O
used NN O O
for NN O O
building NN O O
a NN O O
dictionary NN O I-Generic
: NN O O
an NN O O
English-to-Korean NN O I-Material
dictionary NN O I-Material
and NN O O
English-to-Japanese NN O I-Material
dictionary NN O I-Material
. NN O O

We NN O O
found NN O O
that NN O O
the NN O O
first NN O O
method NN O I-Generic
is NN O O
the NN O O
most NN O O
effective NN O O
and NN O O
the NN O O
best NN O O
result NN O O
can NN O O
be NN O O
obtained NN O O
from NN O O
combining NN O O
the NN O O
three NN O O
methods NN O O
. NN O O

-DOCSTART- NN O O

An NN O O
empirical NN O O
comparison NN O O
of NN O O
CFG NN O I-Method
filtering NN O I-Method
techniques NN O I-Method
for NN O O
LTAG NN O I-Method
and NN O O
HPSG NN O I-Method
is NN O O
presented NN O O
. NN O O

We NN O O
demonstrate NN O O
that NN O O
an NN O O
approximation NN O I-Method
of NN O I-Method
HPSG NN O B-Method
produces NN O O
a NN O O
more NN O O
effective NN O O
CFG NN O I-Method
filter NN O I-Method
than NN O O
that NN O I-Generic
of NN O O
LTAG NN O I-Method
. NN O O

We NN O O
also NN O O
investigate NN O O
the NN O O
reason NN O O
for NN O O
that NN O O
difference NN O O
. NN O O

-DOCSTART- NN O O

Syntax-based NN O I-Task
statistical NN O I-Task
machine NN O I-Task
translation NN O I-Task
-LRB- NN O I-Task
MT NN O I-Task
-RRB- NN O I-Task
aims NN O O
at NN O O
applying NN O O
statistical NN O I-Method
models NN O I-Method
to NN O O
structured NN O I-Material
data NN O I-Material
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
a NN O O
syntax-based NN O I-Task
statistical NN O I-Task
machine NN O I-Task
translation NN O I-Task
system NN O I-Task
based NN O O
on NN O O
a NN O O
probabilistic NN O I-Method
synchronous NN O I-Method
dependency NN O I-Method
insertion NN O I-Method
grammar NN O I-Method
. NN O O

Synchronous NN O I-Method
dependency NN O I-Method
insertion NN O I-Method
grammars NN O I-Method
are NN O O
a NN O O
version NN O O
of NN O O
synchronous NN O I-Method
grammars NN O I-Method
defined NN O O
on NN O O
dependency NN O I-OtherScientificTerm
trees NN O I-OtherScientificTerm
. NN O O

We NN O O
first NN O O
introduce NN O O
our NN O O
approach NN O I-Generic
to NN O O
inducing NN O O
such NN O O
a NN O O
grammar NN O I-Method
from NN O O
parallel NN O I-Material
corpora NN O I-Material
. NN O O

Second NN O O
, NN O O
we NN O O
describe NN O O
the NN O O
graphical NN O I-Method
model NN O I-Method
for NN O O
the NN O O
machine NN O I-Task
translation NN O I-Task
task NN O I-Task
, NN O O
which NN O O
can NN O O
also NN O O
be NN O O
viewed NN O O
as NN O O
a NN O O
stochastic NN O I-Method
tree-to-tree NN O I-Method
transducer NN O I-Method
. NN O O

We NN O O
introduce NN O O
a NN O O
polynomial NN O I-Method
time NN O I-Method
decoding NN O I-Method
algorithm NN O I-Method
for NN O O
the NN O O
model NN O I-Generic
. NN O O

We NN O O
evaluate NN O O
the NN O O
outputs NN O O
of NN O O
our NN O O
MT NN O I-Method
system NN O I-Method
using NN O O
the NN O O
NIST NN O I-Metric
and NN O I-Metric
Bleu NN O I-Metric
automatic NN O I-Metric
MT NN O I-Metric
evaluation NN O I-Metric
software NN O I-Metric
. NN O O

The NN O O
result NN O O
shows NN O O
that NN O O
our NN O O
system NN O I-Generic
outperforms NN O O
the NN O O
baseline NN O I-Generic
system NN O I-Generic
based NN O O
on NN O O
the NN O O
IBM NN O I-Method
models NN O I-Method
in NN O O
both NN O O
translation NN O I-Metric
speed NN O I-Metric
and NN O I-Metric
quality NN O I-Metric
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
a NN O O
framework NN O I-Generic
to NN O O
derive NN O O
the NN O O
distance NN O O
between NN O O
concepts NN O I-OtherScientificTerm
from NN O O
distributional NN O I-Method
measures NN O I-Method
of NN O I-Method
word NN O I-Method
co-occurrences NN O I-Method
. NN O O

We NN O O
use NN O O
the NN O O
categories NN O O
in NN O O
a NN O O
published NN O O
thesaurus NN O O
as NN O O
coarse-grained NN O I-OtherScientificTerm
concepts NN O I-OtherScientificTerm
, NN O O
allowing NN O O
all NN O O
possible NN O O
distance NN O O
values NN O O
to NN O O
be NN O O
stored NN O O
in NN O O
a NN O O
concept-concept NN O I-OtherScientificTerm
matrix NN O I-OtherScientificTerm
roughly NN O O
.01 NN O O
% NN O O
the NN O O
size NN O O
of NN O O
that NN O O
created NN O O
by NN O O
existing NN O O
measures NN O O
. NN O O

We NN O O
show NN O O
that NN O O
the NN O O
newly NN O O
proposed NN O O
concept-distance NN O I-Method
measures NN O I-Method
outperform NN O O
traditional NN O O
distributional NN O I-Method
word-distance NN O I-Method
measures NN O I-Method
in NN O O
the NN O O
tasks NN O I-Generic
of NN O O
-LRB- NN O O
1 NN O O
-RRB- NN O O
ranking NN O I-Task
word NN O I-Task
pairs NN O I-Task
in NN O I-Task
order NN O I-Task
of NN O I-Task
semantic NN O I-Task
distance NN O I-Task
, NN O O
and NN O O
-LRB- NN O O
2 NN O O
-RRB- NN O O
correcting NN O I-Task
real-word NN O I-Task
spelling NN O I-Task
errors NN O I-Task
. NN O O

In NN O O
the NN O O
latter NN O O
task NN O I-Generic
, NN O O
of NN O O
all NN O O
the NN O O
WordNet-based NN O I-Method
measures NN O I-Method
, NN O O
only NN O O
that NN O O
proposed NN O O
by NN O O
Jiang NN O O
and NN O O
Conrath NN O O
outperforms NN O O
the NN O O
best NN O O
distributional NN O I-Method
concept-distance NN O I-Method
measures NN O I-Method
. NN O O

-DOCSTART- NN O O

We NN O O
focus NN O O
on NN O O
the NN O O
problem NN O O
of NN O O
building NN O O
large NN O O
repositories NN O O
of NN O O
lexical NN O I-Method
conceptual NN O I-Method
structure NN O I-Method
-LRB- NN O I-Method
LCS NN O I-Method
-RRB- NN O I-Method
representations NN O I-Method
for NN O O
verbs NN O O
in NN O O
multiple NN O O
languages NN O O
. NN O O

One NN O O
of NN O O
the NN O O
main NN O O
results NN O O
of NN O O
this NN O O
work NN O O
is NN O O
the NN O O
definition NN O O
of NN O O
a NN O O
relation NN O O
between NN O O
broad NN O I-OtherScientificTerm
semantic NN O I-OtherScientificTerm
classes NN O I-OtherScientificTerm
and NN O O
LCS NN O I-Method
meaning NN O I-Method
components NN O I-Method
. NN O O

Our NN O O
acquisition NN O I-Method
program NN O I-Method
- NN O I-Method
LEXICALL NN O I-Method
- NN O I-Method
takes NN O O
, NN O O
as NN O O
input NN O O
, NN O O
the NN O O
result NN O O
of NN O O
previous NN O O
work NN O O
on NN O O
verb NN O I-Task
classification NN O I-Task
and NN O O
thematic NN O I-Task
grid NN O I-Task
tagging NN O I-Task
, NN O O
and NN O O
outputs NN O O
LCS NN O I-Method
representations NN O I-Method
for NN O O
different NN O O
languages NN O O
. NN O O

These NN O O
representations NN O I-Generic
have NN O O
been NN O O
ported NN O O
into NN O O
English NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
Arabic NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
Spanish NN O I-OtherScientificTerm
lexicons NN O I-OtherScientificTerm
, NN O O
each NN O O
containing NN O O
approximately NN O O
9000 NN O O
verbs NN O O
. NN O O

We NN O O
are NN O O
currently NN O O
using NN O O
these NN O O
lexicons NN O I-Generic
in NN O O
an NN O O
operational NN O I-Task
foreign NN O I-Task
language NN O I-Task
tutoring NN O I-Task
and NN O O
machine NN O I-Task
translation NN O I-Task
. NN O O

-DOCSTART- NN O O

The NN O O
theoretical NN O O
study NN O O
of NN O O
the NN O O
range NN O I-Method
concatenation NN O I-Method
grammar NN O I-Method
-LSB- NN O I-Method
RCG NN O I-Method
-RSB- NN O I-Method
formalism NN O I-Method
has NN O O
revealed NN O O
many NN O O
attractive NN O O
properties NN O O
which NN O O
may NN O O
be NN O O
used NN O O
in NN O O
NLP NN O I-Task
. NN O O

In NN O O
particular NN O O
, NN O O
range NN O I-OtherScientificTerm
concatenation NN O I-OtherScientificTerm
languages NN O I-OtherScientificTerm
-LSB- NN O I-OtherScientificTerm
RCL NN O I-OtherScientificTerm
-RSB- NN O I-OtherScientificTerm
can NN O O
be NN O O
parsed NN O O
in NN O O
polynomial NN O I-OtherScientificTerm
time NN O I-OtherScientificTerm
and NN O O
many NN O O
classical NN O O
grammatical NN O I-Method
formalisms NN O I-Method
can NN O O
be NN O O
translated NN O O
into NN O O
equivalent NN O O
RCGs NN O I-Method
without NN O O
increasing NN O O
their NN O O
worst-case NN O I-Metric
parsing NN O I-Metric
time NN O I-Metric
complexity NN O I-Metric
. NN O O

For NN O O
example NN O O
, NN O O
after NN O O
translation NN O O
into NN O O
an NN O O
equivalent NN O O
RCG NN O I-Method
, NN O O
any NN O O
tree NN O I-Method
adjoining NN O I-Method
grammar NN O I-Method
can NN O O
be NN O O
parsed NN O O
in NN O O
O NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
n6 NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
time NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
study NN O O
a NN O O
parsing NN O I-Method
technique NN O I-Method
whose NN O O
purpose NN O O
is NN O O
to NN O O
improve NN O O
the NN O O
practical NN O O
efficiency NN O O
of NN O O
RCL NN O I-Method
parsers NN O I-Method
. NN O O

The NN O O
non-deterministic NN O I-Method
parsing NN O I-Method
choices NN O O
of NN O O
the NN O O
main NN O I-Method
parser NN O I-Method
for NN O O
a NN O O
language NN O I-OtherScientificTerm
L NN O I-OtherScientificTerm
are NN O O
directed NN O O
by NN O O
a NN O O
guide NN O O
which NN O O
uses NN O O
the NN O O
shared NN O I-OtherScientificTerm
derivation NN O I-OtherScientificTerm
forest NN O I-OtherScientificTerm
output NN O O
by NN O O
a NN O O
prior NN O O
RCL NN O I-Method
parser NN O I-Method
for NN O O
a NN O O
suitable NN O O
superset NN O O
of NN O O
L NN O O
. NN O O

The NN O O
results NN O O
of NN O O
a NN O O
practical NN O O
evaluation NN O O
of NN O O
this NN O O
method NN O I-Generic
on NN O O
a NN O O
wide NN O I-Method
coverage NN O I-Method
English NN O I-Method
grammar NN O I-Method
are NN O O
given NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
introduce NN O O
Ant-Q NN O I-Method
, NN O O
a NN O O
family NN O O
of NN O O
algorithms NN O O
which NN O O
present NN O O
many NN O O
similarities NN O O
with NN O O
Q-learning NN O I-Method
-LRB- NN O O
Watkins NN O O
, NN O O
1989 NN O O
-RRB- NN O O
, NN O O
and NN O O
which NN O O
we NN O O
apply NN O O
to NN O O
the NN O O
solution NN O O
of NN O O
symmetric NN O I-Task
and NN O I-Task
asym-metric NN O I-Task
instances NN O I-Task
of NN O I-Task
the NN O I-Task
traveling NN O I-Task
salesman NN O I-Task
problem NN O I-Task
-LRB- NN O I-Task
TSP NN O I-Task
-RRB- NN O I-Task
. NN O O

Ant-Q NN O I-Method
algorithms NN O I-Method
were NN O O
inspired NN O O
by NN O O
work NN O O
on NN O O
the NN O O
ant NN O I-Method
system NN O I-Method
-LRB- NN O I-Method
AS NN O I-Method
-RRB- NN O I-Method
, NN O O
a NN O O
distributed NN O I-Method
algorithm NN O I-Method
for NN O O
combinatorial NN O I-Task
optimization NN O I-Task
based NN O O
on NN O O
the NN O O
metaphor NN O O
of NN O O
ant NN O O
colonies NN O O
which NN O O
was NN O O
recently NN O O
proposed NN O O
in NN O O
-LRB- NN O O
Dorigo NN O O
, NN O O
1992 NN O O
; NN O O
Dorigo NN O O
, NN O O
Maniezzo NN O O
and NN O O
Colorni NN O O
, NN O O
1996 NN O O
-RRB- NN O O
. NN O O

We NN O O
show NN O O
that NN O O
AS NN O I-Method
is NN O O
a NN O O
particular NN O O
instance NN O O
of NN O O
the NN O O
Ant-Q NN O I-Method
family NN O I-Method
, NN O O
and NN O O
that NN O O
there NN O O
are NN O O
instances NN O I-Generic
of NN O O
this NN O O
family NN O I-Generic
which NN O O
perform NN O O
better NN O O
than NN O O
AS NN O I-OtherScientificTerm
. NN O O

We NN O O
experimentally NN O O
investigate NN O O
the NN O O
functioning NN O O
of NN O O
Ant-Q NN O I-Method
and NN O O
we NN O O
show NN O O
that NN O O
the NN O O
results NN O O
obtained NN O O
by NN O O
Ant-Q NN O I-Method
on NN O O
symmetric NN O I-Task
TSP NN O I-Task
's NN O O
are NN O O
competitive NN O O
with NN O O
those NN O O
obtained NN O O
by NN O O
other NN O O
heuristic NN O I-Method
approaches NN O I-Method
based NN O O
on NN O O
neural NN O I-Method
networks NN O I-Method
or NN O O
local NN O I-Method
search NN O I-Method
. NN O O

Finally NN O O
, NN O O
we NN O O
apply NN O O
Ant-Q NN O I-Method
to NN O O
some NN O O
difficult NN O O
asymmetric NN O I-Task
TSP NN O I-Task
's NN O O
obtaining NN O O
very NN O O
good NN O O
results NN O O
: NN O O
Ant-Q NN O I-Method
was NN O O
able NN O O
to NN O O
find NN O O
solutions NN O O
of NN O O
a NN O O
quality NN O O
which NN O O
usually NN O O
can NN O O
be NN O O
found NN O O
only NN O O
by NN O O
very NN O O
specialized NN O O
algorithms NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
develop NN O O
a NN O O
geometric NN O I-Method
framework NN O I-Method
for NN O O
linear NN O I-Task
or NN O I-Task
nonlinear NN O I-Task
discriminant NN O I-Task
subspace NN O I-Task
learning NN O I-Task
and NN O I-Task
classification NN O I-Task
. NN O O

In NN O O
our NN O O
framework NN O I-Generic
, NN O O
the NN O O
structures NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
classes NN O I-OtherScientificTerm
are NN O O
conceptualized NN O O
as NN O O
a NN O O
semi-Riemannian NN O I-OtherScientificTerm
manifold NN O I-OtherScientificTerm
which NN O O
is NN O O
considered NN O O
as NN O O
a NN O O
submanifold NN O I-OtherScientificTerm
embedded NN O O
in NN O O
an NN O O
ambient NN O I-OtherScientificTerm
semi-Riemannian NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
. NN O O

The NN O O
class NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
of NN O O
original NN O O
samples NN O O
can NN O O
be NN O O
characterized NN O O
and NN O O
deformed NN O O
by NN O O
local NN O I-Metric
metrics NN O I-Metric
of NN O I-Metric
the NN O I-Metric
semi-Riemannian NN O I-Metric
space NN O I-Metric
. NN O O

Semi-Riemannian NN O I-Metric
metrics NN O I-Metric
are NN O O
uniquely NN O O
determined NN O O
by NN O O
the NN O O
smoothing NN O I-Method
of NN O I-Method
discrete NN O I-Method
functions NN O I-Method
and NN O O
the NN O O
nullity NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
semi-Riemannian NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
. NN O O

Based NN O O
on NN O O
the NN O O
geometrization NN O I-Method
of NN O I-Method
class NN O I-Method
structures NN O I-Method
, NN O O
optimizing NN O O
class NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
in NN O O
the NN O O
feature NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
is NN O O
equivalent NN O O
to NN O O
maximizing NN O O
the NN O O
quadratic NN O I-OtherScientificTerm
quantities NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
metric NN O I-OtherScientificTerm
tensors NN O I-OtherScientificTerm
in NN O O
the NN O O
semi-Riemannian NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
. NN O O

Thus NN O O
supervised NN O I-Method
discriminant NN O I-Method
subspace NN O I-Method
learning NN O I-Method
reduces NN O O
to NN O O
unsupervised NN O I-Method
semi-Riemannian NN O I-Method
mani-fold NN O I-Method
learning NN O I-Method
. NN O O

Based NN O O
on NN O O
the NN O O
proposed NN O O
framework NN O I-Generic
, NN O O
a NN O O
novel NN O O
algorithm NN O I-Generic
, NN O O
dubbed NN O O
as NN O O
Semi-Riemannian NN O I-Method
Discriminant NN O I-Method
Analysis NN O I-Method
-LRB- NN O I-Method
SRDA NN O I-Method
-RRB- NN O I-Method
, NN O O
is NN O O
presented NN O O
for NN O O
subspace-based NN O I-Task
classification NN O I-Task
. NN O O

The NN O O
performance NN O O
of NN O O
SRDA NN O I-Method
is NN O O
tested NN O O
on NN O O
face NN O I-Task
recognition NN O I-Task
-LRB- NN O I-Task
singular NN O I-Task
case NN O I-Task
-RRB- NN O O
and NN O O
handwritten NN O I-Task
capital NN O I-Task
letter NN O I-Task
classification NN O I-Task
-LRB- NN O I-Task
nonsingular NN O I-Task
case NN O I-Task
-RRB- NN O I-Task
against NN O O
existing NN O O
algorithms NN O I-Generic
. NN O O

The NN O O
experimental NN O O
results NN O O
show NN O O
that NN O O
SRDA NN O I-Generic
works NN O O
well NN O O
on NN O O
recognition NN O I-Task
and NN O O
classification NN O I-Task
, NN O O
implying NN O O
that NN O O
semi-Riemannian NN O I-Method
geometry NN O I-Method
is NN O O
a NN O O
promising NN O O
new NN O O
tool NN O O
for NN O O
pattern NN O I-Task
recognition NN O I-Task
and NN O O
machine NN O I-Task
learning NN O I-Task
. NN O O

-DOCSTART- NN O O

A NN O O
deterministic NN O I-Method
parser NN O I-Method
is NN O O
under NN O O
development NN O O
which NN O O
represents NN O O
a NN O O
departure NN O O
from NN O O
traditional NN O O
deterministic NN O I-Method
parsers NN O I-Method
in NN O O
that NN O O
it NN O I-Generic
combines NN O O
both NN O O
symbolic NN O I-Method
and NN O I-Method
connectionist NN O I-Method
components NN O I-Method
. NN O O

The NN O O
connectionist NN O I-Method
component NN O I-Method
is NN O O
trained NN O O
either NN O O
from NN O O
patterns NN O I-OtherScientificTerm
derived NN O O
from NN O O
the NN O O
rules NN O I-Method
of NN O I-Method
a NN O I-Method
deterministic NN O I-Method
grammar NN O I-Method
. NN O O

The NN O O
development NN O O
and NN O O
evolution NN O O
of NN O O
such NN O O
a NN O O
hybrid NN O I-Method
architecture NN O I-Method
has NN O O
lead NN O O
to NN O O
a NN O O
parser NN O I-Method
which NN O O
is NN O O
superior NN O O
to NN O O
any NN O O
known NN O O
deterministic NN O I-Method
parser NN O I-Method
. NN O O

Experiments NN O O
are NN O O
described NN O O
and NN O O
powerful NN O O
training NN O I-Method
techniques NN O I-Method
are NN O O
demonstrated NN O O
that NN O O
permit NN O O
decision-making NN O I-Task
by NN O O
the NN O O
connectionist NN O I-Method
component NN O I-Method
in NN O O
the NN O O
parsing NN O I-Method
process NN O I-Method
. NN O O

This NN O O
approach NN O I-Generic
has NN O O
permitted NN O O
some NN O O
simplifications NN O O
to NN O O
the NN O O
rules NN O I-OtherScientificTerm
of NN O O
other NN O O
deterministic NN O I-Method
parsers NN O I-Method
, NN O O
including NN O O
the NN O O
elimination NN O O
of NN O O
rule NN O O
packets NN O O
and NN O O
priorities NN O O
. NN O O

Furthermore NN O O
, NN O O
parsing NN O I-Task
is NN O O
performed NN O O
more NN O O
robustly NN O O
and NN O O
with NN O O
more NN O O
tolerance NN O O
for NN O O
error NN O O
. NN O O

Data NN O O
are NN O O
presented NN O O
which NN O O
show NN O O
how NN O O
a NN O O
connectionist NN O I-Method
-LRB- NN O I-Method
neural NN O I-Method
-RRB- NN O I-Method
network NN O I-Method
trained NN O O
with NN O O
linguistic NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
can NN O O
parse NN O O
both NN O O
expected NN O I-Material
-LRB- NN O I-Material
grammatical NN O I-Material
-RRB- NN O I-Material
sentences NN O I-Material
as NN O O
well NN O O
as NN O O
some NN O O
novel NN O O
-LRB- NN O I-Material
ungrammatical NN O I-Material
or NN O I-Material
lexically NN O I-Material
ambiguous NN O I-Material
-RRB- NN O I-Material
sentences NN O I-Material
. NN O O

-DOCSTART- NN O O

Robust NN O O
natural NN O I-Task
language NN O I-Task
interpretation NN O I-Task
requires NN O O
strong NN O O
semantic NN O I-Method
domain NN O I-Method
models NN O I-Method
, NN O O
fail-soft NN O I-Method
recovery NN O I-Method
heuristics NN O I-Method
, NN O O
and NN O O
very NN O O
flexible NN O O
control NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
. NN O O

Although NN O O
single-strategy NN O I-Method
parsers NN O I-Method
have NN O O
met NN O O
with NN O O
a NN O O
measure NN O O
of NN O O
success NN O O
, NN O O
a NN O O
multi-strategy NN O I-Method
approach NN O I-Method
is NN O O
shown NN O O
to NN O O
provide NN O O
a NN O O
much NN O O
higher NN O O
degree NN O O
of NN O O
flexibility NN O O
, NN O O
redundancy NN O O
, NN O O
and NN O O
ability NN O O
to NN O O
bring NN O O
task-specific NN O I-OtherScientificTerm
domain NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
-LRB- NN O O
in NN O O
addition NN O O
to NN O O
general NN O I-OtherScientificTerm
linguistic NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
-RRB- NN O O
to NN O O
bear NN O O
on NN O O
both NN O O
grammatical NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
ungrammatical NN O I-OtherScientificTerm
input NN O I-OtherScientificTerm
. NN O O

A NN O O
parsing NN O I-Method
algorithm NN O I-Method
is NN O O
presented NN O O
that NN O O
integrates NN O O
several NN O O
different NN O O
parsing NN O I-Method
strategies NN O I-Method
, NN O O
with NN O O
case-frame NN O I-OtherScientificTerm
instantiation NN O I-OtherScientificTerm
dominating NN O O
. NN O O

Each NN O O
of NN O O
these NN O O
parsing NN O I-Method
strategies NN O I-Method
exploits NN O O
different NN O O
types NN O O
of NN O O
knowledge NN O O
; NN O O
and NN O O
their NN O O
combination NN O O
provides NN O O
a NN O O
strong NN O O
framework NN O O
in NN O O
which NN O O
to NN O O
process NN O O
conjunctions NN O I-OtherScientificTerm
, NN O O
fragmentary NN O I-OtherScientificTerm
input NN O I-OtherScientificTerm
, NN O O
and NN O O
ungrammatical NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
, NN O O
as NN O O
well NN O O
as NN O O
less NN O O
exotic NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
grammatically NN O I-OtherScientificTerm
correct NN O I-OtherScientificTerm
input NN O I-OtherScientificTerm
. NN O O

Several NN O O
specific NN O I-Method
heuristics NN O I-Method
for NN O O
handling NN O O
ungrammatical NN O I-OtherScientificTerm
input NN O I-OtherScientificTerm
are NN O O
presented NN O O
within NN O O
this NN O O
multi-strategy NN O I-Method
framework NN O I-Method
. NN O O

-DOCSTART- NN O O

Recently NN O O
, NN O O
Stacked NN O I-Method
Auto-Encoders NN O I-Method
-LRB- NN O I-Method
SAE NN O I-Method
-RRB- NN O I-Method
have NN O O
been NN O O
successfully NN O O
used NN O O
for NN O O
learning NN O I-Task
imbalanced NN O I-Task
datasets NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
for NN O O
the NN O O
first NN O O
time NN O O
, NN O O
we NN O O
propose NN O O
to NN O O
use NN O O
a NN O O
Neural NN O I-Method
Network NN O I-Method
classifier NN O I-Method
furnished NN O O
by NN O O
an NN O O
SAE NN O I-Method
structure NN O I-Method
for NN O O
detecting NN O O
the NN O O
errors NN O O
made NN O O
by NN O O
a NN O O
strong NN O O
Automatic NN O I-Method
Speech NN O I-Method
Recognition NN O I-Method
-LRB- NN O I-Method
ASR NN O I-Method
-RRB- NN O I-Method
system NN O I-Method
. NN O O

Error NN O I-Task
detection NN O I-Task
on NN O O
an NN O O
automatic NN O I-Material
transcription NN O I-Material
provided NN O O
by NN O O
a NN O O
'' NN O O
strong NN O O
'' NN O O
ASR NN O I-Method
system NN O I-Method
, NN O O
i.e. NN O O
exhibiting NN O O
a NN O O
small NN O O
word NN O I-Metric
error NN O I-Metric
rate NN O I-Metric
, NN O O
is NN O O
difficult NN O O
due NN O O
to NN O O
the NN O O
limited NN O O
number NN O O
of NN O O
'' NN O O
positive NN O O
'' NN O O
examples NN O O
-LRB- NN O O
i.e. NN O O
words NN O O
erroneously NN O O
recognized NN O O
-RRB- NN O O
available NN O O
for NN O O
training NN O O
a NN O O
binary NN O I-Method
classi-fier NN O I-Method
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
investigate NN O O
and NN O O
compare NN O O
different NN O O
types NN O O
of NN O O
classifiers NN O I-Method
for NN O O
automatically NN O I-Task
detecting NN O I-Task
ASR NN O I-Task
errors NN O I-Task
, NN O O
including NN O O
the NN O O
one NN O I-Generic
based NN O O
on NN O O
a NN O O
stacked NN O I-Method
auto-encoder NN O I-Method
architecture NN O I-Method
. NN O O

We NN O O
show NN O O
the NN O O
effectiveness NN O O
of NN O O
the NN O O
latter NN O O
by NN O O
measuring NN O O
and NN O O
comparing NN O O
performance NN O O
on NN O O
the NN O O
automatic NN O I-Material
transcriptions NN O I-Material
of NN O O
an NN O O
English NN O I-Material
corpus NN O I-Material
collected NN O O
from NN O O
TED NN O I-Material
talks NN O I-Material
. NN O O

Performance NN O O
of NN O O
each NN O O
investigated NN O O
classifier NN O I-Method
is NN O O
evaluated NN O O
both NN O O
via NN O O
receiving NN O I-Metric
operating NN O I-Metric
curve NN O I-Metric
and NN O O
via NN O O
a NN O O
measure NN O I-Generic
, NN O O
called NN O O
mean NN O I-Metric
absolute NN O I-Metric
error NN O I-Metric
, NN O O
related NN O O
to NN O O
the NN O O
quality NN O O
in NN O O
predicting NN O O
the NN O O
corresponding NN O O
word NN O I-Metric
error NN O I-Metric
rate NN O I-Metric
. NN O O

The NN O O
results NN O O
demonstrates NN O O
that NN O O
the NN O O
classifier NN O I-Method
based NN O O
on NN O O
SAE NN O I-Method
detects NN O O
the NN O O
ASR NN O I-OtherScientificTerm
errors NN O I-OtherScientificTerm
better NN O O
than NN O O
the NN O O
other NN O O
classification NN O I-Method
methods NN O I-Method
. NN O O

-DOCSTART- NN O O

Within NN O O
the NN O O
EU NN O O
Network NN O O
of NN O O
Excellence NN O O
PASCAL NN O O
, NN O O
a NN O O
challenge NN O O
was NN O O
organized NN O O
to NN O O
design NN O O
a NN O O
statistical NN O I-Method
machine NN O I-Method
learning NN O I-Method
algorithm NN O I-Method
that NN O O
segments NN O O
words NN O O
into NN O O
the NN O O
smallest NN O I-OtherScientificTerm
meaning-bearing NN O I-OtherScientificTerm
units NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
, NN O O
morphemes NN O I-OtherScientificTerm
. NN O O

Ideally NN O O
, NN O O
these NN O I-Generic
are NN O O
basic NN O O
vocabulary NN O O
units NN O O
suitable NN O O
for NN O O
different NN O O
tasks NN O I-Generic
, NN O O
such NN O O
as NN O O
speech NN O I-Task
and NN O I-Task
text NN O I-Task
understanding NN O I-Task
, NN O O
machine NN O I-Task
translation NN O I-Task
, NN O O
information NN O I-Task
retrieval NN O I-Task
, NN O O
and NN O O
statistical NN O I-Task
language NN O I-Task
modeling NN O I-Task
. NN O O

Twelve NN O O
research NN O O
groups NN O O
participated NN O O
in NN O O
the NN O O
challenge NN O O
and NN O O
had NN O O
submitted NN O O
segmentation NN O O
results NN O O
obtained NN O O
by NN O O
their NN O O
algorithms NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
evaluate NN O O
the NN O O
application NN O O
of NN O O
these NN O O
segmen-tation NN O I-Method
algorithms NN O I-Method
to NN O O
large NN O I-Task
vocabulary NN O I-Task
speech NN O I-Task
recognition NN O I-Task
using NN O O
statistical NN O I-Method
n-gram NN O I-Method
language NN O I-Method
models NN O I-Method
based NN O O
on NN O O
the NN O O
proposed NN O O
word NN O O
segments NN O O
instead NN O O
of NN O O
entire NN O O
words NN O O
. NN O O

Experiments NN O O
were NN O O
done NN O O
for NN O O
two NN O O
ag-glutinative NN O I-Material
and NN O I-Material
morphologically NN O I-Material
rich NN O I-Material
languages NN O I-Material
: NN O O
Finnish NN O I-Material
and NN O O
Turk-ish NN O I-Material
. NN O O

We NN O O
also NN O O
investigate NN O O
combining NN O O
various NN O O
segmentations NN O O
to NN O O
improve NN O O
the NN O O
performance NN O O
of NN O O
the NN O O
recognizer NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
a NN O O
recently NN O O
collected NN O O
spoken NN O I-Material
language NN O I-Material
corpus NN O I-Material
for NN O O
the NN O O
ATIS NN O I-Task
-LRB- NN O I-Task
Air NN O I-Task
Travel NN O I-Task
Information NN O I-Task
System NN O I-Task
-RRB- NN O I-Task
domain NN O I-Task
. NN O O

This NN O O
data NN O I-Generic
collection NN O I-Generic
effort NN O O
has NN O O
been NN O O
co-ordinated NN O O
by NN O O
MADCOW NN O O
-LRB- NN O O
Multi-site NN O O
ATIS NN O O
Data NN O O
COllection NN O O
Working NN O O
group NN O O
-RRB- NN O O
. NN O O

We NN O O
summarize NN O O
the NN O O
motivation NN O O
for NN O O
this NN O O
effort NN O O
, NN O O
the NN O O
goals NN O O
, NN O O
the NN O O
implementation NN O O
of NN O O
a NN O O
multi-site NN O I-Method
data NN O I-Method
collection NN O I-Method
paradigm NN O I-Method
, NN O O
and NN O O
the NN O O
accomplishments NN O O
of NN O O
MADCOW NN O O
in NN O O
monitoring NN O O
the NN O O
collection NN O I-Generic
and NN O O
distribution NN O O
of NN O O
12,000 NN O O
utterances NN O O
of NN O O
spontaneous NN O I-Material
speech NN O I-Material
from NN O O
five NN O O
sites NN O O
for NN O O
use NN O O
in NN O O
a NN O O
multi-site NN O I-Task
common NN O I-Task
evaluation NN O I-Task
of NN O I-Task
speech NN O I-Task
, NN O I-Task
natural NN O I-Task
language NN O I-Task
and NN O I-Task
spoken NN O I-Task
language NN O I-Task
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
the NN O O
Hierarchical NN O I-Method
Directed NN O I-Method
Acyclic NN O I-Method
Graph NN O I-Method
-LRB- NN O I-Method
HDAG NN O I-Method
-RRB- NN O I-Method
Kernel NN O I-Method
for NN O O
structured NN O I-Material
natural NN O I-Material
language NN O I-Material
data NN O I-Material
. NN O O

The NN O O
HDAG NN O I-Method
Kernel NN O I-Method
directly NN O O
accepts NN O O
several NN O O
levels NN O O
of NN O O
both NN O O
chunks NN O O
and NN O O
their NN O O
relations NN O O
, NN O O
and NN O O
then NN O O
efficiently NN O O
computes NN O O
the NN O O
weighed NN O O
sum NN O O
of NN O O
the NN O O
number NN O O
of NN O O
common NN O O
attribute NN O O
sequences NN O O
of NN O O
the NN O O
HDAGs NN O I-Method
. NN O O

We NN O O
applied NN O O
the NN O O
proposed NN O O
method NN O I-Generic
to NN O O
question NN O I-Task
classification NN O I-Task
and NN O I-Task
sentence NN O I-Task
alignment NN O I-Task
tasks NN O I-Task
to NN O O
evaluate NN O O
its NN O O
performance NN O O
as NN O O
a NN O O
similarity NN O I-Metric
measure NN O I-Metric
and NN O O
a NN O O
kernel NN O I-Method
function NN O I-Method
. NN O O

The NN O O
results NN O O
of NN O O
the NN O O
experiments NN O O
demonstrate NN O O
that NN O O
the NN O O
HDAG NN O I-Method
Kernel NN O I-Method
is NN O O
superior NN O O
to NN O O
other NN O O
kernel NN O I-Method
functions NN O I-Method
and NN O O
baseline NN O I-Generic
methods NN O I-Generic
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
a NN O O
solution NN O O
to NN O O
the NN O O
challenge NN O O
of NN O O
the NN O O
CoNLL NN O I-Task
2008 NN O I-Task
shared NN O I-Task
task NN O I-Task
that NN O O
uses NN O O
a NN O O
generative NN O I-Method
history-based NN O I-Method
latent NN O I-Method
variable NN O I-Method
model NN O I-Method
to NN O O
predict NN O O
the NN O O
most NN O O
likely NN O O
derivation NN O O
of NN O O
a NN O O
synchronous NN O I-Task
dependency NN O I-Task
parser NN O I-Task
for NN O O
both NN O O
syntactic NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
semantic NN O I-OtherScientificTerm
dependencies NN O I-OtherScientificTerm
. NN O O

The NN O O
submitted NN O O
model NN O I-Generic
yields NN O O
79.1 NN O O
% NN O O
macro-average NN O I-Metric
F1 NN O I-Metric
performance NN O I-Metric
, NN O O
for NN O O
the NN O O
joint NN O O
task NN O I-Generic
, NN O O
86.9 NN O O
% NN O O
syntactic NN O I-Metric
dependencies NN O I-Metric
LAS NN O I-Metric
and NN O O
71.0 NN O O
% NN O O
semantic NN O I-Metric
dependencies NN O I-Metric
F1 NN O I-Metric
. NN O O

A NN O O
larger NN O O
model NN O I-Generic
trained NN O O
after NN O O
the NN O O
deadline NN O O
achieves NN O O
80.5 NN O O
% NN O O
macro-average NN O I-Metric
F1 NN O I-Metric
, NN O O
87.6 NN O O
% NN O O
syntactic NN O I-Metric
dependencies NN O I-Metric
LAS NN O I-Metric
, NN O O
and NN O O
73.1 NN O O
% NN O O
semantic NN O I-Metric
dependencies NN O I-Metric
F1 NN O I-Metric
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
a NN O O
new NN O O
, NN O O
large NN O I-Task
scale NN O I-Task
discourse-level NN O I-Task
annotation NN O I-Task
project NN O O
- NN O O
the NN O O
Penn NN O I-Material
Discourse NN O I-Material
TreeBank NN O I-Material
-LRB- NN O I-Material
PDTB NN O I-Material
-RRB- NN O I-Material
. NN O O

We NN O O
present NN O O
an NN O O
approach NN O I-Generic
to NN O O
annotating NN O O
a NN O O
level NN O O
of NN O O
discourse NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
that NN O O
is NN O O
based NN O O
on NN O O
identifying NN O O
discourse NN O I-OtherScientificTerm
connectives NN O I-OtherScientificTerm
and NN O O
their NN O O
arguments NN O O
. NN O O

The NN O O
PDTB NN O I-Material
is NN O O
being NN O O
built NN O O
directly NN O O
on NN O O
top NN O O
of NN O O
the NN O O
Penn NN O I-Material
TreeBank NN O I-Material
and NN O O
Propbank NN O I-Material
, NN O O
thus NN O O
supporting NN O O
the NN O O
extraction NN O I-Task
of NN O I-Task
useful NN O I-Task
syntactic NN O I-Task
and NN O I-Task
semantic NN O I-Task
features NN O I-Task
and NN O O
providing NN O O
a NN O O
richer NN O O
substrate NN O O
for NN O O
the NN O O
development NN O O
and NN O O
evaluation NN O O
of NN O O
practical NN O I-Method
algorithms NN O I-Method
. NN O O

We NN O O
provide NN O O
a NN O O
detailed NN O O
preliminary NN O O
analysis NN O O
of NN O O
inter-annotator NN O I-Metric
agreement NN O I-Metric
- NN O O
both NN O O
the NN O O
level NN O I-Metric
of NN O I-Metric
agreement NN O I-Metric
and NN O O
the NN O O
types NN O O
of NN O O
inter-annotator NN O I-OtherScientificTerm
variation NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Currently NN O O
, NN O O
N-gram NN O I-Method
models NN O I-Method
are NN O O
the NN O O
most NN O O
common NN O O
and NN O O
widely NN O O
used NN O O
models NN O O
for NN O O
statistical NN O I-Task
language NN O I-Task
modeling NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
investigated NN O O
an NN O O
alternative NN O O
way NN O O
to NN O O
build NN O O
language NN O I-Method
models NN O I-Method
, NN O O
i.e. NN O O
, NN O O
using NN O O
artificial NN O I-Method
neural NN O I-Method
networks NN O I-Method
to NN O O
learn NN O O
the NN O O
language NN O I-Method
model NN O I-Method
. NN O O

Our NN O O
experiment NN O O
result NN O O
shows NN O O
that NN O O
the NN O O
neural NN O I-Method
network NN O I-Method
can NN O O
learn NN O O
a NN O O
language NN O I-Method
model NN O I-Method
that NN O O
has NN O O
performance NN O O
even NN O O
better NN O O
than NN O O
standard NN O O
statistical NN O I-Method
methods NN O I-Method
. NN O O

-DOCSTART- NN O O

Automated NN O I-Task
facial NN O I-Task
expression NN O I-Task
recognition NN O I-Task
has NN O O
received NN O O
increased NN O O
attention NN O O
over NN O O
the NN O O
past NN O O
two NN O O
decades NN O O
. NN O O

Existing NN O I-Generic
works NN O I-Generic
in NN O O
the NN O O
field NN O O
usually NN O O
do NN O O
not NN O O
encode NN O O
either NN O O
the NN O O
temporal NN O I-OtherScientificTerm
evolution NN O I-OtherScientificTerm
or NN O O
the NN O O
intensity NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
observed NN O I-OtherScientificTerm
facial NN O I-OtherScientificTerm
displays NN O I-OtherScientificTerm
. NN O O

They NN O I-Generic
also NN O O
fail NN O O
to NN O O
jointly NN O O
model NN O O
multidimensional NN O I-Material
-LRB- NN O I-Material
multi-class NN O I-Material
-RRB- NN O I-Material
continuous NN O I-Material
facial NN O I-Material
behaviour NN O I-Material
data NN O I-Material
; NN O O
binary NN O O
classifiers-one NN O O
for NN O O
each NN O O
target NN O O
basic-emotion NN O O
class-are NN O O
used NN O O
instead NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
intrinsic NN O I-OtherScientificTerm
topology NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
multidimensional NN O I-OtherScientificTerm
continuous NN O I-OtherScientificTerm
facial NN O I-OtherScientificTerm
affect NN O O
data NN O O
is NN O O
first NN O O
modeled NN O O
by NN O O
an NN O O
ordinal NN O I-OtherScientificTerm
man-ifold NN O I-OtherScientificTerm
. NN O O

This NN O O
topology NN O I-Generic
is NN O O
then NN O O
incorporated NN O O
into NN O O
the NN O O
Hidden NN O I-Method
Conditional NN O I-Method
Ordinal NN O I-Method
Random NN O I-Method
Field NN O I-Method
-LRB- NN O I-Method
H-CORF NN O I-Method
-RRB- NN O I-Method
framework NN O I-Method
for NN O O
dynamic NN O I-Task
ordinal NN O I-Task
regression NN O I-Task
by NN O O
constraining NN O O
H-CORF NN O I-OtherScientificTerm
parameters NN O I-OtherScientificTerm
to NN O O
lie NN O O
on NN O O
the NN O O
ordinal NN O I-OtherScientificTerm
manifold NN O I-OtherScientificTerm
. NN O O

The NN O O
resulting NN O O
model NN O I-Generic
attains NN O O
simultaneous NN O I-Task
dynamic NN O I-Task
recognition NN O I-Task
and NN O O
intensity NN O I-Task
estimation NN O I-Task
of NN O I-Task
facial NN O I-Task
expressions NN O I-Task
of NN O O
multiple NN O O
emotions NN O O
. NN O O

To NN O O
the NN O O
best NN O O
of NN O O
our NN O O
knowledge NN O O
, NN O O
the NN O I-Method
proposed NN O I-Method
method NN O I-Method
is NN O O
the NN O O
first NN O O
one NN O O
to NN O O
achieve NN O O
this NN O O
on NN O O
both NN O O
deliberate NN O O
as NN O O
well NN O O
as NN O O
spontaneous NN O I-Material
facial NN O I-Material
affect NN O I-Material
data NN O I-Material
. NN O O

-DOCSTART- NN O O

Recent NN O O
advances NN O O
in NN O O
linear NN O I-Task
classification NN O I-Task
have NN O O
shown NN O O
that NN O O
for NN O O
applications NN O I-Generic
such NN O O
as NN O O
document NN O I-Task
classification NN O I-Task
, NN O O
the NN O O
training NN O O
can NN O O
be NN O O
extremely NN O O
efficient NN O O
. NN O O

However NN O O
, NN O O
most NN O O
of NN O O
the NN O O
existing NN O O
training NN O I-Method
methods NN O I-Method
are NN O O
designed NN O O
by NN O O
assuming NN O O
that NN O O
data NN O O
can NN O O
be NN O O
stored NN O O
in NN O O
the NN O O
computer NN O I-OtherScientificTerm
memory NN O I-OtherScientificTerm
. NN O O

These NN O O
methods NN O I-Generic
can NN O O
not NN O O
be NN O O
easily NN O O
applied NN O O
to NN O O
data NN O I-Generic
larger NN O O
than NN O O
the NN O O
memory NN O I-OtherScientificTerm
capacity NN O I-OtherScientificTerm
due NN O O
to NN O O
the NN O O
random NN O I-OtherScientificTerm
access NN O I-OtherScientificTerm
to NN O O
the NN O O
disk NN O I-OtherScientificTerm
. NN O O

We NN O O
propose NN O O
and NN O O
analyze NN O O
a NN O O
block NN O I-Method
minimization NN O I-Method
framework NN O I-Method
for NN O O
data NN O I-Generic
larger NN O O
than NN O O
the NN O O
memory NN O I-OtherScientificTerm
size NN O I-OtherScientificTerm
. NN O O

At NN O O
each NN O O
step NN O O
a NN O O
block NN O O
of NN O O
data NN O O
is NN O O
loaded NN O O
from NN O O
the NN O O
disk NN O I-OtherScientificTerm
and NN O O
handled NN O O
by NN O O
certain NN O O
learning NN O I-Method
methods NN O I-Method
. NN O O

We NN O O
investigate NN O O
two NN O O
implementations NN O O
of NN O O
the NN O O
proposed NN O O
framework NN O I-Generic
for NN O O
primal NN O I-Method
and NN O I-Method
dual NN O I-Method
SVMs NN O I-Method
, NN O O
respectively NN O O
. NN O O

As NN O O
data NN O O
can NN O O
not NN O O
fit NN O O
in NN O O
memory NN O O
, NN O O
many NN O O
design NN O O
considerations NN O O
are NN O O
very NN O O
different NN O O
from NN O O
those NN O O
for NN O O
traditional NN O O
algorithms NN O O
. NN O O

Experiments NN O O
using NN O O
data NN O O
sets NN O O
20 NN O O
times NN O O
larger NN O O
than NN O O
the NN O O
memory NN O O
demonstrate NN O O
the NN O O
effectiveness NN O O
of NN O O
the NN O O
proposed NN O O
method NN O I-Generic
. NN O O

-DOCSTART- NN O O

Instances NN O O
of NN O O
a NN O O
word NN O O
drawn NN O O
from NN O O
different NN O O
domains NN O O
may NN O O
have NN O O
different NN O O
sense NN O I-OtherScientificTerm
priors NN O I-OtherScientificTerm
-LRB- NN O O
the NN O O
proportions NN O O
of NN O O
the NN O O
different NN O O
senses NN O O
of NN O O
a NN O O
word NN O O
-RRB- NN O O
. NN O O

This NN O O
in NN O O
turn NN O O
affects NN O O
the NN O O
accuracy NN O I-Metric
of NN O O
word NN O I-Method
sense NN O I-Method
disambiguation NN O I-Method
-LRB- NN O I-Method
WSD NN O I-Method
-RRB- NN O I-Method
systems NN O I-Method
trained NN O O
and NN O O
applied NN O O
on NN O O
different NN O O
domains NN O O
. NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
method NN O I-Generic
to NN O O
estimate NN O O
the NN O O
sense NN O I-OtherScientificTerm
priors NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
drawn NN O O
from NN O O
a NN O O
new NN O I-Material
domain NN O I-Material
, NN O O
and NN O O
highlights NN O O
the NN O O
importance NN O O
of NN O O
using NN O O
well NN O I-OtherScientificTerm
calibrated NN O I-OtherScientificTerm
probabilities NN O I-OtherScientificTerm
when NN O O
performing NN O O
these NN O O
estimations NN O I-Generic
. NN O O

By NN O O
using NN O O
well NN O I-OtherScientificTerm
calibrated NN O I-OtherScientificTerm
probabilities NN O I-OtherScientificTerm
, NN O O
we NN O O
are NN O O
able NN O O
to NN O O
estimate NN O O
the NN O O
sense NN O I-OtherScientificTerm
priors NN O I-OtherScientificTerm
effectively NN O O
to NN O O
achieve NN O O
significant NN O O
improvements NN O O
in NN O O
WSD NN O I-Metric
accuracy NN O I-Metric
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
deal NN O O
with NN O O
a NN O O
recently NN O O
developed NN O O
large NN O I-Material
Czech NN O I-Material
MWE NN O I-Material
database NN O I-Material
containing NN O O
at NN O O
the NN O O
moment NN O O
160 NN O O
000 NN O O
MWEs NN O I-OtherScientificTerm
-LRB- NN O O
treated NN O O
as NN O O
lexical NN O I-OtherScientificTerm
units NN O I-OtherScientificTerm
-RRB- NN O O
. NN O O

It NN O I-Generic
was NN O O
compiled NN O O
from NN O O
various NN O O
resources NN O O
such NN O O
as NN O O
encyclopedias NN O I-Material
and NN O O
dictionaries NN O I-Material
, NN O O
public NN O I-Material
databases NN O I-Material
of NN O I-Material
proper NN O I-Material
names NN O I-Material
and NN O I-Material
toponyms NN O I-Material
, NN O O
collocations NN O I-Generic
obtained NN O O
from NN O O
Czech NN O I-Material
WordNet NN O I-Material
, NN O O
lists NN O I-Material
of NN O I-Material
botanical NN O I-Material
and NN O I-Material
zoological NN O I-Material
terms NN O I-Material
and NN O O
others NN O O
. NN O O

We NN O O
describe NN O O
the NN O O
structure NN O O
of NN O O
the NN O O
database NN O I-Generic
and NN O O
give NN O O
basic NN O O
types NN O O
of NN O O
MWEs NN O I-OtherScientificTerm
according NN O O
to NN O O
domains NN O O
they NN O O
belong NN O O
to NN O O
. NN O O

We NN O O
compare NN O O
the NN O O
built NN O O
MWEs NN O I-Material
database NN O I-Material
with NN O O
the NN O O
corpus NN O O
data NN O O
from NN O O
Czech NN O I-Material
National NN O I-Material
Corpus NN O I-Material
-LRB- NN O O
approx NN O O
. NN O O

100 NN O O
mil NN O O
. NN O O

tokens NN O O
-RRB- NN O O
and NN O O
present NN O O
results NN O O
of NN O O
this NN O O
comparison NN O O
in NN O O
the NN O O
paper NN O O
. NN O O

These NN O O
MWEs NN O I-OtherScientificTerm
have NN O O
not NN O O
been NN O O
obtained NN O O
from NN O O
the NN O O
corpus NN O I-Generic
since NN O O
their NN O O
frequencies NN O O
in NN O O
it NN O O
are NN O O
rather NN O O
low NN O O
. NN O O

To NN O O
obtain NN O O
a NN O O
more NN O O
complete NN O O
list NN O O
of NN O O
MWEs NN O I-OtherScientificTerm
we NN O O
propose NN O O
and NN O O
use NN O O
a NN O O
technique NN O I-Generic
exploiting NN O O
the NN O O
Word NN O I-Task
Sketch NN O I-Task
Engine NN O I-Task
, NN O O
which NN O O
allows NN O O
us NN O O
to NN O O
work NN O O
with NN O O
statistical NN O I-OtherScientificTerm
parameters NN O I-OtherScientificTerm
such NN O O
as NN O O
frequency NN O O
of NN O O
MWEs NN O I-OtherScientificTerm
and NN O O
their NN O O
components NN O O
as NN O O
well NN O O
as NN O O
with NN O O
the NN O O
salience NN O O
for NN O O
the NN O O
whole NN O O
MWEs NN O I-OtherScientificTerm
. NN O O

We NN O O
also NN O O
discuss NN O O
exploitation NN O O
of NN O O
the NN O O
database NN O I-Generic
for NN O O
working NN O O
out NN O O
a NN O O
more NN O O
adequate NN O O
tagging NN O I-Task
and NN O O
lemmatization NN O I-Task
. NN O O

The NN O O
final NN O O
goal NN O O
is NN O O
to NN O O
be NN O O
able NN O O
to NN O O
recognize NN O O
MWEs NN O I-OtherScientificTerm
in NN O O
corpus NN O O
text NN O O
and NN O O
lemmatize NN O O
them NN O I-Generic
as NN O O
complete NN O O
lexical NN O I-OtherScientificTerm
units NN O I-OtherScientificTerm
, NN O O
i. NN O O
e. NN O O
to NN O O
make NN O O
tagging NN O I-Task
and NN O O
lemmatization NN O I-Task
more NN O O
adequate NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
the NN O O
ongoing NN O O
construction NN O O
of NN O O
a NN O O
large NN O O
, NN O O
semantically NN O I-Material
annotated NN O I-Material
corpus NN O I-Material
resource NN O O
as NN O O
reliable NN O O
basis NN O O
for NN O O
the NN O O
large-scale NN O I-Task
acquisition NN O I-Task
of NN O I-Task
word-semantic NN O I-Task
information NN O I-Task
, NN O O
e.g. NN O O
the NN O O
construction NN O I-Task
of NN O I-Task
domain-independent NN O I-Task
lexica NN O I-Task
. NN O O

The NN O O
backbone NN O O
of NN O O
the NN O O
annotation NN O I-OtherScientificTerm
are NN O O
semantic NN O I-OtherScientificTerm
roles NN O I-OtherScientificTerm
in NN O O
the NN O O
frame NN O I-Method
semantics NN O I-Method
paradigm NN O I-Method
. NN O O

We NN O O
report NN O O
experiences NN O O
and NN O O
evaluate NN O O
the NN O O
annotated NN O I-Material
data NN O I-Material
from NN O O
the NN O O
first NN O O
project NN O O
stage NN O O
. NN O O

On NN O O
this NN O O
basis NN O O
, NN O O
we NN O O
discuss NN O O
the NN O O
problems NN O O
of NN O O
vagueness NN O I-OtherScientificTerm
and NN O O
ambiguity NN O I-OtherScientificTerm
in NN O O
semantic NN O I-Method
annotation NN O I-Method
. NN O O

-DOCSTART- NN O O

Statistical NN O I-Method
machine NN O I-Method
translation NN O I-Method
-LRB- NN O I-Method
SMT NN O I-Method
-RRB- NN O I-Method
is NN O O
currently NN O O
one NN O O
of NN O O
the NN O O
hot NN O O
spots NN O O
in NN O O
natural NN O I-Task
language NN O I-Task
processing NN O I-Task
. NN O O

Over NN O O
the NN O O
last NN O O
few NN O O
years NN O O
dramatic NN O O
improvements NN O O
have NN O O
been NN O O
made NN O O
, NN O O
and NN O O
a NN O O
number NN O O
of NN O O
comparative NN O O
evaluations NN O O
have NN O O
shown NN O O
, NN O O
that NN O O
SMT NN O I-Method
gives NN O O
competitive NN O O
results NN O O
to NN O O
rule-based NN O I-Method
translation NN O I-Method
systems NN O I-Method
, NN O O
requiring NN O O
significantly NN O O
less NN O O
development NN O O
time NN O O
. NN O O

This NN O O
is NN O O
particularly NN O O
important NN O O
when NN O O
building NN O O
translation NN O I-Task
systems NN O I-Task
for NN O O
new NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
pairs NN O I-OtherScientificTerm
or NN O O
new NN O I-OtherScientificTerm
domains NN O I-OtherScientificTerm
. NN O O

This NN O O
workshop NN O O
is NN O O
intended NN O O
to NN O O
give NN O O
an NN O O
introduction NN O O
to NN O O
statistical NN O I-Method
machine NN O I-Method
translation NN O I-Method
with NN O O
a NN O O
focus NN O O
on NN O O
practical NN O O
considerations NN O O
. NN O O

Participants NN O O
should NN O O
be NN O O
able NN O O
, NN O O
after NN O O
attending NN O O
this NN O O
workshop NN O O
, NN O O
to NN O O
set NN O O
out NN O O
building NN O O
an NN O O
SMT NN O I-Method
system NN O I-Method
themselves NN O O
and NN O O
achieving NN O O
good NN O O
baseline NN O O
results NN O O
in NN O O
a NN O O
short NN O O
time NN O O
. NN O O

The NN O O
tutorial NN O O
will NN O O
cover NN O O
the NN O O
basics NN O O
of NN O O
SMT NN O I-Method
: NN O O
Theory NN O O
will NN O O
be NN O O
put NN O O
into NN O O
practice NN O O
. NN O O

STTK NN O I-Method
, NN O O
a NN O O
statistical NN O I-Method
machine NN O I-Method
translation NN O I-Method
tool NN O I-Method
kit NN O I-Method
, NN O O
will NN O O
be NN O O
introduced NN O O
and NN O O
used NN O O
to NN O O
build NN O O
a NN O O
working NN O O
translation NN O I-Task
system NN O I-Task
. NN O O

STTK NN O I-Method
has NN O O
been NN O O
developed NN O O
by NN O O
the NN O O
presenter NN O O
and NN O O
co-workers NN O O
over NN O O
a NN O O
number NN O O
of NN O O
years NN O O
and NN O O
is NN O O
currently NN O O
used NN O O
as NN O O
the NN O O
basis NN O O
of NN O O
CMU NN O O
's NN O O
SMT NN O I-Method
system NN O I-Method
. NN O O

It NN O I-Generic
has NN O O
also NN O O
successfully NN O O
been NN O O
coupled NN O O
with NN O O
rule-based NN O I-Method
and NN O I-Method
example NN O I-Method
based NN O I-Method
machine NN O I-Method
translation NN O I-Method
modules NN O I-Method
to NN O O
build NN O O
a NN O O
multi NN O I-Task
engine NN O I-Task
machine NN O I-Task
translation NN O I-Task
system NN O I-Task
. NN O O

The NN O O
source NN O O
code NN O O
of NN O O
the NN O O
tool NN O I-Generic
kit NN O I-Generic
will NN O O
be NN O O
made NN O O
available NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
an NN O O
unsupervised NN O I-Method
learning NN O I-Method
approach NN O I-Method
to NN O O
building NN O O
a NN O O
non-English NN O I-Task
-LRB- NN O I-Task
Arabic NN O I-Task
-RRB- NN O I-Task
stemmer NN O I-Task
. NN O O

The NN O O
stemming NN O I-Method
model NN O I-Method
is NN O O
based NN O O
on NN O O
statistical NN O I-Method
machine NN O I-Method
translation NN O I-Method
and NN O O
it NN O I-Generic
uses NN O O
an NN O O
English NN O I-Method
stemmer NN O I-Method
and NN O O
a NN O O
small NN O O
-LRB- NN O O
10K NN O O
sentences NN O O
-RRB- NN O O
parallel NN O I-Material
corpus NN O I-Material
as NN O O
its NN O O
sole NN O O
training NN O O
resources NN O O
. NN O O

No NN O O
parallel NN O I-Material
text NN O I-Material
is NN O O
needed NN O O
after NN O O
the NN O O
training NN O O
phase NN O O
. NN O O

Monolingual NN O I-Material
, NN O I-Material
unannotated NN O I-Material
text NN O I-Material
can NN O O
be NN O O
used NN O O
to NN O O
further NN O O
improve NN O O
the NN O O
stemmer NN O I-Method
by NN O O
allowing NN O O
it NN O I-Generic
to NN O O
adapt NN O O
to NN O O
a NN O O
desired NN O O
domain NN O O
or NN O O
genre NN O O
. NN O O

Examples NN O O
and NN O O
results NN O O
will NN O O
be NN O O
given NN O O
for NN O O
Arabic NN O I-Material
, NN O O
but NN O O
the NN O O
approach NN O I-Generic
is NN O O
applicable NN O O
to NN O O
any NN O O
language NN O O
that NN O O
needs NN O O
affix NN O I-OtherScientificTerm
removal NN O I-OtherScientificTerm
. NN O O

Our NN O O
resource-frugal NN O I-Method
approach NN O I-Method
results NN O O
in NN O O
87.5 NN O O
% NN O O
agreement NN O I-Metric
with NN O O
a NN O O
state NN O O
of NN O O
the NN O O
art NN O O
, NN O O
proprietary NN O O
Arabic NN O I-Method
stemmer NN O I-Method
built NN O O
using NN O O
rules NN O I-OtherScientificTerm
, NN O O
affix NN O I-Material
lists NN O I-Material
, NN O O
and NN O O
human NN O I-Material
annotated NN O I-Material
text NN O I-Material
, NN O O
in NN O O
addition NN O O
to NN O O
an NN O O
unsupervised NN O I-Method
component NN O I-Method
. NN O O

Task-based NN O I-Method
evaluation NN O I-Method
using NN O O
Arabic NN O I-Task
information NN O I-Task
retrieval NN O I-Task
indicates NN O O
an NN O O
improvement NN O O
of NN O O
22-38 NN O O
% NN O O
in NN O O
average NN O I-Metric
precision NN O I-Metric
over NN O O
unstemmed NN O I-Material
text NN O I-Material
, NN O O
and NN O O
96 NN O O
% NN O O
of NN O O
the NN O O
performance NN O O
of NN O O
the NN O O
proprietary NN O O
stemmer NN O I-Method
above NN O O
. NN O O

-DOCSTART- NN O O

The NN O O
paper NN O O
assesses NN O O
the NN O O
capability NN O O
of NN O O
an NN O O
HMM-based NN O I-Method
TTS NN O I-Method
system NN O I-Method
to NN O O
produce NN O O
German NN O I-Material
speech NN O I-Material
. NN O O

The NN O O
results NN O O
are NN O O
discussed NN O O
in NN O O
qualitative NN O O
terms NN O O
, NN O O
and NN O O
compared NN O O
over NN O O
three NN O O
different NN O O
choices NN O O
of NN O O
context NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
. NN O O

In NN O O
addition NN O O
, NN O O
the NN O O
system NN O I-Generic
is NN O O
adapted NN O O
to NN O O
a NN O O
small NN O O
set NN O O
of NN O O
football NN O I-Material
announcements NN O I-Material
, NN O O
in NN O O
an NN O O
exploratory NN O O
attempt NN O O
to NN O O
synthe-sise NN O O
expressive NN O I-Material
speech NN O I-Material
. NN O O

We NN O O
conclude NN O O
that NN O O
the NN O O
HMMs NN O I-Method
are NN O O
able NN O O
to NN O O
produce NN O O
highly NN O O
intelligible NN O I-Material
neutral NN O I-Material
German NN O I-Material
speech NN O I-Material
, NN O O
with NN O O
a NN O O
stable NN O O
quality NN O O
, NN O O
and NN O O
that NN O O
the NN O O
expressivity NN O O
is NN O O
partially NN O O
captured NN O O
in NN O O
spite NN O O
of NN O O
the NN O O
small NN O O
size NN O O
of NN O O
the NN O O
football NN O I-Material
dataset NN O I-Material
. NN O O

-DOCSTART- NN O O

We NN O O
give NN O O
an NN O O
analysis NN O I-Task
of NN O I-Task
ellipsis NN O I-Task
resolution NN O I-Task
in NN O O
terms NN O O
of NN O O
a NN O O
straightforward NN O O
discourse NN O I-Method
copying NN O I-Method
algorithm NN O I-Method
that NN O O
correctly NN O O
predicts NN O O
a NN O O
wide NN O O
range NN O O
of NN O O
phenomena NN O O
. NN O O

The NN O O
treatment NN O I-Generic
does NN O O
not NN O O
suffer NN O O
from NN O O
problems NN O O
inherent NN O O
in NN O O
identity-of-relations NN O I-Task
analyses NN O I-Task
. NN O O

Furthermore NN O O
, NN O O
in NN O O
contrast NN O O
to NN O O
the NN O O
approach NN O O
of NN O O
Dalrymple NN O O
et NN O O
al. NN O O
-LSB- NN O O
1991 NN O O
-RSB- NN O O
, NN O O
the NN O O
treatment NN O I-Generic
directly NN O O
encodes NN O O
the NN O O
intuitive NN O O
distinction NN O O
between NN O O
full NN O I-OtherScientificTerm
NPs NN O I-OtherScientificTerm
and NN O O
the NN O O
referential NN O I-OtherScientificTerm
elements NN O I-OtherScientificTerm
that NN O O
corefer NN O O
with NN O O
them NN O O
through NN O O
what NN O O
we NN O O
term NN O O
role NN O I-Task
linking NN O I-Task
. NN O O

The NN O O
correct NN O O
predictions NN O O
for NN O O
several NN O O
problematic NN O O
examples NN O O
of NN O O
ellipsis NN O I-OtherScientificTerm
naturally NN O O
result NN O O
. NN O O

Finally NN O O
, NN O O
the NN O O
analysis NN O I-Generic
extends NN O O
directly NN O O
to NN O O
other NN O O
discourse NN O I-Task
copying NN O I-Task
phenomena NN O I-Task
. NN O O

-DOCSTART- NN O O

How NN O O
to NN O O
obtain NN O O
hierarchical NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
-LRB- NN O O
e.g. NN O O
superordinate NN O I-OtherScientificTerm
- NN O I-OtherScientificTerm
hyponym NN O I-OtherScientificTerm
relation NN O I-OtherScientificTerm
, NN O O
synonym NN O I-OtherScientificTerm
relation NN O I-OtherScientificTerm
-RRB- NN O O
is NN O O
one NN O O
of NN O O
the NN O O
most NN O O
important NN O O
problems NN O O
for NN O O
thesaurus NN O I-Task
construction NN O I-Task
. NN O O

A NN O O
pilot NN O O
system NN O O
for NN O O
extracting NN O O
these NN O O
relations NN O I-Generic
automatically NN O O
from NN O O
an NN O O
ordinary NN O O
Japanese NN O I-Material
language NN O I-Material
dictionary NN O I-Material
-LRB- NN O O
Shinmeikai NN O O
Kokugojiten NN O O
, NN O O
published NN O O
by NN O O
Sansei-do NN O O
, NN O O
in NN O O
machine NN O O
readable NN O O
form NN O O
-RRB- NN O O
is NN O O
given NN O O
. NN O O

The NN O O
features NN O I-OtherScientificTerm
of NN O O
the NN O O
definition NN O I-OtherScientificTerm
sentences NN O I-OtherScientificTerm
in NN O O
the NN O O
dictionary NN O I-Generic
, NN O O
the NN O O
mechanical NN O O
extraction NN O O
of NN O O
the NN O O
hierarchical NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
and NN O O
the NN O O
estimation NN O O
of NN O O
the NN O O
results NN O O
are NN O O
discussed NN O O
. NN O O

-DOCSTART- NN O O

Face NN O I-Task
images NN O I-Task
of NN O I-Task
non-frontal NN O I-Task
views NN O I-Task
under NN O O
poor NN O O
illumination NN O O
with NN O O
low NN O O
resolution NN O O
reduce NN O O
dramatically NN O O
face NN O I-Metric
recognition NN O I-Metric
accuracy NN O I-Metric
. NN O O

This NN O O
is NN O O
evident NN O O
most NN O O
compellingly NN O O
by NN O O
the NN O O
very NN O O
low NN O O
recognition NN O I-Metric
rate NN O I-Metric
of NN O O
all NN O O
existing NN O O
face NN O I-Method
recognition NN O I-Method
systems NN O I-Method
when NN O O
applied NN O O
to NN O O
live NN O I-Material
CCTV NN O I-Material
camera NN O I-Material
input NN O I-Material
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
a NN O O
Bayesian NN O I-Method
framework NN O I-Method
to NN O O
perform NN O O
multi-modal NN O O
-LRB- NN O O
such NN O O
as NN O O
variations NN O O
in NN O O
viewpoint NN O I-OtherScientificTerm
and NN O O
illumination NN O I-OtherScientificTerm
-RRB- NN O O
face NN O I-Task
image NN O I-Task
super-resolution NN O I-Task
for NN O O
recognition NN O I-Task
in NN O O
tensor NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
. NN O O

Given NN O O
a NN O O
single NN O I-Material
modal NN O I-Material
low-resolution NN O I-Material
face NN O I-Material
image NN O I-Material
, NN O O
we NN O O
benefit NN O O
from NN O O
the NN O O
multiple NN O I-OtherScientificTerm
factor NN O I-OtherScientificTerm
interactions NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
training NN O I-OtherScientificTerm
tensor NN O I-OtherScientificTerm
, NN O O
and NN O O
super-resolve NN O O
its NN O O
high-resolution NN O I-Task
reconstructions NN O I-Task
across NN O O
different NN O O
modalities NN O I-OtherScientificTerm
for NN O O
face NN O I-Task
recognition NN O I-Task
. NN O O

Instead NN O O
of NN O O
performing NN O O
pixel-domain NN O I-Task
super-resolution NN O I-Task
and NN O I-Task
recognition NN O I-Task
independently NN O O
as NN O O
two NN O O
separate NN O O
sequential NN O O
processes NN O O
, NN O O
we NN O O
integrate NN O O
the NN O O
tasks NN O O
of NN O O
super-resolution NN O I-Task
and NN O O
recognition NN O I-Task
by NN O O
directly NN O O
computing NN O O
a NN O O
maximum NN O I-OtherScientificTerm
likelihood NN O I-OtherScientificTerm
identity NN O I-OtherScientificTerm
parameter NN O I-OtherScientificTerm
vector NN O I-OtherScientificTerm
in NN O O
high-resolution NN O I-OtherScientificTerm
tensor NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
for NN O O
recognition NN O I-Task
. NN O O

We NN O O
show NN O O
results NN O O
from NN O O
multi-modal NN O I-Task
super-resolution NN O I-Task
and NN O I-Task
face NN O I-Task
recognition NN O I-Task
experiments NN O O
across NN O O
different NN O O
imaging NN O I-OtherScientificTerm
modalities NN O I-OtherScientificTerm
, NN O O
using NN O O
low-resolution NN O I-Material
images NN O I-Material
as NN O O
testing NN O O
inputs NN O O
and NN O O
demonstrate NN O O
improved NN O O
recognition NN O I-Metric
rates NN O I-Metric
over NN O O
standard NN O O
tensorface NN O I-Method
and NN O I-Method
eigenface NN O I-Method
representations NN O I-Method
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
describe NN O O
a NN O O
phrase-based NN O I-Method
unigram NN O I-Method
model NN O I-Method
for NN O O
statistical NN O I-Task
machine NN O I-Task
translation NN O I-Task
that NN O O
uses NN O O
a NN O O
much NN O O
simpler NN O O
set NN O O
of NN O O
model NN O I-OtherScientificTerm
parameters NN O I-OtherScientificTerm
than NN O O
similar NN O O
phrase-based NN O I-Method
models NN O I-Method
. NN O O

The NN O O
units NN O O
of NN O O
translation NN O O
are NN O O
blocks NN O I-OtherScientificTerm
- NN O O
pairs NN O O
of NN O O
phrases NN O O
. NN O O

During NN O O
decoding NN O I-Task
, NN O O
we NN O O
use NN O O
a NN O O
block NN O I-Method
unigram NN O I-Method
model NN O I-Method
and NN O O
a NN O O
word-based NN O I-Method
trigram NN O I-Method
language NN O I-Method
model NN O I-Method
. NN O O

During NN O O
training NN O I-Task
, NN O O
the NN O O
blocks NN O I-OtherScientificTerm
are NN O O
learned NN O O
from NN O O
source NN O I-Method
interval NN O I-Method
projections NN O I-Method
using NN O O
an NN O O
underlying NN O O
word NN O I-Method
alignment NN O I-Method
. NN O O

We NN O O
show NN O O
experimental NN O O
results NN O O
on NN O O
block NN O I-Metric
selection NN O I-Metric
criteria NN O I-Metric
based NN O O
on NN O O
unigram NN O I-OtherScientificTerm
counts NN O I-OtherScientificTerm
and NN O O
phrase NN O I-OtherScientificTerm
length NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
develops NN O O
a NN O O
new NN O O
approach NN O I-Generic
for NN O O
extremely NN O O
fast NN O I-Task
detection NN O I-Task
in NN O O
domains NN O O
where NN O O
the NN O O
distribution NN O I-OtherScientificTerm
of NN O O
positive NN O I-Material
and NN O I-Material
negative NN O I-Material
examples NN O I-Material
is NN O O
highly NN O O
skewed NN O O
-LRB- NN O O
e.g. NN O O
face NN O I-Task
detection NN O I-Task
or NN O O
database NN O I-Task
retrieval NN O I-Task
-RRB- NN O O
. NN O O

In NN O O
such NN O O
domains NN O O
a NN O O
cascade NN O I-Method
of NN O I-Method
simple NN O I-Method
classifiers NN O B-Method
each NN O O
trained NN O O
to NN O O
achieve NN O O
high NN O O
detection NN O I-Metric
rates NN O I-Metric
and NN O O
modest NN O I-Metric
false NN O I-Metric
positive NN O I-Metric
rates NN O I-Metric
can NN O O
yield NN O O
a NN O O
final NN O O
detector NN O I-Generic
with NN O O
many NN O O
desirable NN O O
features NN O I-OtherScientificTerm
: NN O O
including NN O O
high NN O O
detection NN O I-Metric
rates NN O I-Metric
, NN O O
very NN O O
low NN O O
false NN O I-Metric
positive NN O I-Metric
rates NN O I-Metric
, NN O O
and NN O O
fast NN O I-Metric
performance NN O I-Metric
. NN O O

Achieving NN O O
extremely NN O O
high NN O O
detection NN O I-Metric
rates NN O I-Metric
, NN O O
rather NN O O
than NN O O
low NN O I-Metric
error NN O I-Metric
, NN O O
is NN O O
not NN O O
a NN O O
task NN O O
typically NN O O
addressed NN O O
by NN O O
machine NN O I-Method
learning NN O I-Method
algorithms NN O I-Method
. NN O O

We NN O O
propose NN O O
a NN O O
new NN O O
variant NN O O
of NN O O
AdaBoost NN O I-Method
as NN O O
a NN O O
mechanism NN O O
for NN O O
training NN O O
the NN O O
simple NN O O
classifiers NN O I-OtherScientificTerm
used NN O O
in NN O O
the NN O O
cascade NN O I-Generic
. NN O O

Experimental NN O O
results NN O O
in NN O O
the NN O O
domain NN O O
of NN O O
face NN O I-Task
detection NN O I-Task
show NN O O
the NN O O
training NN O I-Method
algorithm NN O I-Method
yields NN O O
significant NN O O
improvements NN O O
in NN O O
performance NN O O
over NN O O
conventional NN O O
AdaBoost NN O I-Method
. NN O O

The NN O O
final NN O O
face NN O I-Method
detection NN O I-Method
system NN O I-Method
can NN O O
process NN O O
15 NN O O
frames NN O O
per NN O O
second NN O O
, NN O O
achieves NN O O
over NN O O
90 NN O O
% NN O O
detection NN O I-Metric
, NN O O
and NN O O
a NN O O
false NN O I-Metric
positive NN O I-Metric
rate NN O I-Metric
of NN O O
1 NN O O
in NN O O
a NN O O
1,000,000 NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
a NN O O
method NN O I-Generic
for NN O O
learning NN O O
joint NN O I-OtherScientificTerm
embed-dings NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
images NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
text NN O I-OtherScientificTerm
using NN O O
a NN O O
two-branch NN O I-Method
neural NN O I-Method
network NN O I-Method
with NN O O
multiple NN O I-OtherScientificTerm
layers NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
linear NN O I-OtherScientificTerm
projections NN O I-OtherScientificTerm
followed NN O O
by NN O O
nonlinearities NN O I-OtherScientificTerm
. NN O O

The NN O O
network NN O I-Generic
is NN O O
trained NN O O
using NN O O
a NN O O
large-margin NN O I-OtherScientificTerm
objective NN O I-OtherScientificTerm
that NN O O
combines NN O O
cross-view NN O I-OtherScientificTerm
ranking NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
with NN O O
within-view NN O I-OtherScientificTerm
neighborhood NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
preservation NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
inspired NN O O
by NN O O
metric NN O I-OtherScientificTerm
learning NN O I-OtherScientificTerm
literature NN O I-OtherScientificTerm
. NN O O

Extensive NN O O
experiments NN O O
show NN O O
that NN O O
our NN O O
approach NN O I-Generic
gains NN O O
significant NN O O
improvements NN O O
in NN O O
accuracy NN O I-Metric
for NN O O
image-to-text NN O I-Task
and NN O I-Task
text-to-image NN O I-Task
retrieval NN O I-Task
. NN O O

Our NN O O
method NN O I-Generic
achieves NN O O
new NN O O
state-of-the-art NN O O
results NN O O
on NN O O
the NN O O
Flickr30K NN O I-Material
and NN O I-Material
MSCOCO NN O I-Material
image-sentence NN O I-Material
datasets NN O I-Material
and NN O O
shows NN O O
promise NN O O
on NN O O
the NN O O
new NN O O
task NN O O
of NN O O
phrase NN O I-Task
lo-calization NN O I-Task
on NN O O
the NN O O
Flickr30K NN O I-Material
Entities NN O I-Material
dataset NN O I-Material
. NN O O

-DOCSTART- NN O O

State-of-the-art NN O O
Question NN O I-Method
Answering NN O I-Method
-LRB- NN O I-Method
QA NN O I-Method
-RRB- NN O I-Method
systems NN O I-Method
are NN O O
very NN O O
sensitive NN O O
to NN O O
variations NN O O
in NN O O
the NN O O
phrasing NN O O
of NN O O
an NN O O
information NN O O
need NN O O
. NN O O

Finding NN O O
the NN O O
preferred NN O O
language NN O O
for NN O O
such NN O O
a NN O O
need NN O O
is NN O O
a NN O O
valuable NN O O
task NN O O
. NN O O

We NN O O
investigate NN O O
that NN O O
claim NN O O
by NN O O
adopting NN O O
a NN O O
simple NN O O
MT-based NN O I-Method
paraphrasing NN O I-Method
technique NN O I-Method
and NN O O
evaluating NN O O
QA NN O I-Method
system NN O I-Method
performance NN O O
on NN O O
paraphrased NN O I-Material
questions NN O I-Material
. NN O O

We NN O O
found NN O O
a NN O O
potential NN O O
increase NN O O
of NN O O
35 NN O O
% NN O O
in NN O O
MRR NN O I-Metric
with NN O O
respect NN O O
to NN O O
the NN O O
original NN O O
question NN O O
. NN O O

-DOCSTART- NN O O

The NN O O
TAP-XL NN O I-Task
Automated NN O I-Task
Analyst NN O I-Task
's NN O I-Task
Assistant NN O I-Task
is NN O O
an NN O O
application NN O O
designed NN O O
to NN O O
help NN O O
an NN O O
English NN O O
- NN O O
speaking NN O O
analyst NN O O
write NN O O
a NN O O
topical NN O O
report NN O O
, NN O O
culling NN O O
information NN O O
from NN O O
a NN O O
large NN O O
inflow NN O O
of NN O O
multilingual NN O I-Material
, NN O I-Material
multimedia NN O I-Material
data NN O I-Material
. NN O O

It NN O I-Generic
gives NN O O
users NN O O
the NN O O
ability NN O O
to NN O O
spend NN O O
their NN O O
time NN O O
finding NN O O
more NN O O
data NN O O
relevant NN O O
to NN O O
their NN O O
task NN O O
, NN O O
and NN O O
gives NN O O
them NN O O
translingual NN O O
reach NN O O
into NN O O
other NN O O
languages NN O O
by NN O O
leveraging NN O O
human NN O I-Method
language NN O I-Method
technology NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
discusses NN O O
the NN O O
application NN O O
of NN O O
Unification NN O I-Method
Categorial NN O I-Method
Grammar NN O I-Method
-LRB- NN O I-Method
UCG NN O I-Method
-RRB- NN O I-Method
to NN O O
the NN O O
framework NN O O
of NN O O
Isomorphic NN O I-Method
Grammars NN O I-Method
for NN O O
Machine NN O I-Task
Translation NN O I-Task
pioneered NN O O
by NN O O
Landsbergen NN O O
. NN O O

The NN O O
Isomorphic NN O I-Method
Grammars NN O I-Method
approach NN O I-Method
to NN O O
MT NN O I-Task
involves NN O O
developing NN O O
the NN O O
grammars NN O O
of NN O O
the NN O O
Source NN O O
and NN O O
Target NN O O
languages NN O O
in NN O O
parallel NN O O
, NN O O
in NN O O
order NN O O
to NN O O
ensure NN O O
that NN O O
SL NN O O
and NN O O
TL NN O O
expressions NN O O
which NN O O
stand NN O O
in NN O O
the NN O O
translation NN O I-OtherScientificTerm
relation NN O I-OtherScientificTerm
have NN O O
isomorphic NN O I-OtherScientificTerm
derivations NN O I-OtherScientificTerm
. NN O O

The NN O O
principle NN O O
advantage NN O O
of NN O O
this NN O O
approach NN O O
is NN O O
that NN O O
knowledge NN O O
concerning NN O O
translation NN O O
equivalence NN O O
of NN O O
expressions NN O O
may NN O O
be NN O O
directly NN O O
exploited NN O O
, NN O O
obviating NN O O
the NN O O
need NN O O
for NN O O
answers NN O O
to NN O O
semantic NN O O
questions NN O O
that NN O O
we NN O O
do NN O O
not NN O O
yet NN O O
have NN O O
. NN O O

Semantic NN O O
and NN O O
other NN O O
information NN O O
may NN O O
still NN O O
be NN O O
incorporated NN O O
, NN O O
but NN O O
as NN O O
constraints NN O O
on NN O O
the NN O O
translation NN O I-OtherScientificTerm
relation NN O I-OtherScientificTerm
, NN O O
not NN O O
as NN O O
levels NN O O
of NN O O
textual NN O I-Method
representation NN O I-Method
. NN O O

After NN O O
introducing NN O O
this NN O O
approach NN O I-Generic
to NN O O
MT NN O I-Task
system NN O I-Task
design NN O I-Task
, NN O O
and NN O O
the NN O O
basics NN O O
of NN O O
monolingual NN O I-Task
UCG NN O I-Task
, NN O O
we NN O O
will NN O O
show NN O O
how NN O O
the NN O O
two NN O I-Generic
can NN O O
be NN O O
integrated NN O O
, NN O O
and NN O O
present NN O O
an NN O O
example NN O O
from NN O O
an NN O O
implemented NN O O
bi-directional NN O I-OtherScientificTerm
English-Spanish NN O I-OtherScientificTerm
fragment NN O I-OtherScientificTerm
. NN O O

Finally NN O O
we NN O O
will NN O O
present NN O O
some NN O O
outstanding NN O O
problems NN O O
with NN O O
the NN O O
approach NN O I-Generic
. NN O O

-DOCSTART- NN O O

In NN O O
the NN O O
security NN O I-Task
domain NN O I-Task
a NN O O
key NN O O
problem NN O O
is NN O O
identifying NN O I-Task
rare NN O I-OtherScientificTerm
behaviours NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
interest NN O I-OtherScientificTerm
. NN O O

Training NN O I-Material
examples NN O I-Material
for NN O O
these NN O O
behaviours NN O I-Generic
may NN O O
or NN O O
may NN O O
not NN O O
exist NN O O
, NN O O
and NN O O
if NN O O
they NN O I-Generic
do NN O O
exist NN O O
there NN O O
will NN O O
be NN O O
few NN O O
examples NN O O
, NN O O
quite NN O O
probably NN O O
one NN O O
. NN O O

We NN O O
present NN O O
a NN O O
novel NN O O
weakly NN O I-Method
supervised NN O I-Method
algorithm NN O I-Method
that NN O O
can NN O O
detect NN O O
behaviours NN O I-OtherScientificTerm
that NN O O
either NN O O
have NN O O
never NN O O
before NN O O
been NN O O
seen NN O O
or NN O O
for NN O O
which NN O O
there NN O O
are NN O O
few NN O O
examples NN O O
. NN O O

Global NN O I-OtherScientificTerm
context NN O I-OtherScientificTerm
is NN O O
modelled NN O O
, NN O O
allowing NN O O
the NN O O
detection NN O I-Task
of NN O I-Task
abnormal NN O I-Task
behaviours NN O I-Task
that NN O O
in NN O O
isolation NN O O
appear NN O O
normal NN O O
. NN O O

Pragmatic NN O I-OtherScientificTerm
aspects NN O I-OtherScientificTerm
are NN O O
considered NN O O
, NN O O
such NN O O
that NN O O
no NN O O
parameter NN O I-Method
tuning NN O I-Method
is NN O O
required NN O O
and NN O O
real NN O O
time NN O O
performance NN O O
is NN O O
achieved NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
have NN O O
developed NN O O
a NN O O
computational NN O I-Generic
model NN O I-Generic
of NN O O
the NN O O
process NN O O
of NN O O
describing NN O O
the NN O O
layout NN O O
of NN O O
an NN O O
apartment NN O O
or NN O O
house NN O O
, NN O O
a NN O O
much-studied NN O O
discourse NN O I-Task
task NN O I-Task
first NN O O
characterized NN O O
linguistically NN O O
by NN O O
Linde NN O O
-LRB- NN O O
1974 NN O O
-RRB- NN O O
. NN O O

The NN O O
model NN O I-Generic
is NN O O
embodied NN O O
in NN O O
a NN O O
program NN O I-Generic
, NN O O
APT NN O I-Method
, NN O O
that NN O O
can NN O O
reproduce NN O O
segments NN O O
of NN O O
actual NN O O
tape-recorded NN O O
descriptions NN O O
, NN O O
using NN O O
organizational NN O I-Method
and NN O I-Method
discourse NN O I-Method
strategies NN O I-Method
derived NN O O
through NN O O
analysis NN O O
of NN O O
our NN O O
corpus NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
a NN O O
practical NN O O
approach NN O I-Generic
employing NN O O
n-gram NN O I-Method
models NN O I-Method
and NN O O
error-correction NN O I-Method
rules NN O I-Method
for NN O O
Thai NN O I-Task
key NN O I-Task
prediction NN O I-Task
and NN O O
Thai-English NN O I-Task
language NN O I-Task
identification NN O I-Task
. NN O O

The NN O O
paper NN O O
also NN O O
proposes NN O O
rule-reduction NN O I-Method
algorithm NN O I-Method
applying NN O O
mutual NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
to NN O O
reduce NN O O
the NN O O
error-correction NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
. NN O O

Our NN O O
algorithm NN O I-Generic
reported NN O O
more NN O O
than NN O O
99 NN O O
% NN O O
accuracy NN O I-Metric
in NN O O
both NN O O
language NN O I-Task
identification NN O I-Task
and NN O O
key NN O I-Task
prediction NN O I-Task
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
concerns NN O O
the NN O O
discourse NN O I-Task
understanding NN O I-Task
process NN O I-Task
in NN O O
spoken NN O I-Method
dialogue NN O I-Method
systems NN O I-Method
. NN O O

This NN O O
process NN O O
enables NN O O
the NN O O
system NN O I-Generic
to NN O O
understand NN O O
user NN O I-Material
utterances NN O I-Material
based NN O O
on NN O O
the NN O O
context NN O O
of NN O O
a NN O O
dialogue NN O O
. NN O O

Since NN O O
multiple NN O O
candidates NN O O
for NN O O
the NN O O
understanding NN O O
result NN O O
can NN O O
be NN O O
obtained NN O O
for NN O O
a NN O O
user NN O I-Material
utterance NN O I-Material
due NN O O
to NN O O
the NN O O
ambiguity NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
speech NN O I-OtherScientificTerm
understanding NN O I-OtherScientificTerm
, NN O O
it NN O O
is NN O O
not NN O O
appropriate NN O O
to NN O O
decide NN O O
on NN O O
a NN O O
single NN O O
understanding NN O O
result NN O O
after NN O O
each NN O O
user NN O I-OtherScientificTerm
utterance NN O I-OtherScientificTerm
. NN O O

By NN O O
holding NN O O
multiple NN O O
candidates NN O O
for NN O O
understanding NN O O
results NN O O
and NN O O
resolving NN O O
the NN O O
ambiguity NN O I-Generic
as NN O O
the NN O O
dialogue NN O O
progresses NN O O
, NN O O
the NN O O
discourse NN O I-Metric
understanding NN O I-Metric
accuracy NN O I-Metric
can NN O O
be NN O O
improved NN O O
. NN O O

This NN O O
paper NN O O
proposes NN O O
a NN O O
method NN O I-Generic
for NN O O
resolving NN O O
this NN O O
ambiguity NN O I-Generic
based NN O O
on NN O O
statistical NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
obtained NN O O
from NN O O
dialogue NN O I-Material
corpora NN O I-Material
. NN O O

Unlike NN O O
conventional NN O O
methods NN O I-Generic
that NN O O
use NN O O
hand-crafted NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
, NN O O
the NN O O
proposed NN O O
method NN O I-Generic
enables NN O O
easy NN O O
design NN O O
of NN O O
the NN O O
discourse NN O I-Task
understanding NN O I-Task
process NN O I-Task
. NN O O

Experiment NN O O
results NN O O
have NN O O
shown NN O O
that NN O O
a NN O O
system NN O I-Generic
that NN O O
exploits NN O O
the NN O O
proposed NN O O
method NN O I-Generic
performs NN O O
sufficiently NN O O
and NN O O
that NN O O
holding NN O O
multiple NN O O
candidates NN O O
for NN O O
understanding NN O O
results NN O O
is NN O O
effective NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
consider NN O O
the NN O O
problem NN O O
of NN O O
question-focused NN O I-Task
sentence NN O I-Task
retrieval NN O I-Task
from NN O O
complex NN O O
news NN O I-Material
articles NN O I-Material
describing NN O O
multi-event NN O I-Material
stories NN O I-Material
published NN O O
over NN O O
time NN O O
. NN O O

Annotators NN O O
generated NN O O
a NN O O
list NN O O
of NN O O
questions NN O O
central NN O O
to NN O O
understanding NN O O
each NN O O
story NN O O
in NN O O
our NN O O
corpus NN O O
. NN O O

Because NN O O
of NN O O
the NN O O
dynamic NN O O
nature NN O O
of NN O O
the NN O O
stories NN O O
, NN O O
many NN O O
questions NN O O
are NN O O
time-sensitive NN O O
-LRB- NN O O
e.g. NN O O
`` NN O O
How NN O O
many NN O O
victims NN O O
have NN O O
been NN O O
found NN O O
? NN O O
'' NN O O
-RRB- NN O O

. NN O O

Judges NN O O
found NN O O
sentences NN O O
providing NN O O
an NN O O
answer NN O O
to NN O O
each NN O O
question NN O O
. NN O O

To NN O O
address NN O O
the NN O O
sentence NN O I-Task
retrieval NN O I-Task
problem NN O I-Task
, NN O O
we NN O O
apply NN O O
a NN O O
stochastic NN O I-Method
, NN O I-Method
graph-based NN O I-Method
method NN O I-Method
for NN O O
comparing NN O O
the NN O O
relative NN O O
importance NN O O
of NN O O
the NN O O
textual NN O O
units NN O O
, NN O O
which NN O O
was NN O O
previously NN O O
used NN O O
successfully NN O O
for NN O O
generic NN O I-Task
summarization NN O I-Task
. NN O O

Currently NN O O
, NN O O
we NN O O
present NN O O
a NN O O
topic-sensitive NN O O
version NN O O
of NN O O
our NN O O
method NN O I-Generic
and NN O O
hypothesize NN O O
that NN O O
it NN O I-Generic
can NN O O
outperform NN O O
a NN O O
competitive NN O O
baseline NN O I-Generic
, NN O O
which NN O O
compares NN O O
the NN O O
similarity NN O O
of NN O O
each NN O O
sentence NN O O
to NN O O
the NN O O
input NN O O
question NN O O
via NN O O
IDF-weighted NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
overlap NN O I-OtherScientificTerm
. NN O O

In NN O O
our NN O O
experiments NN O O
, NN O O
the NN O O
method NN O I-Generic
achieves NN O O
a NN O O
TRDR NN O I-Metric
score NN O I-Metric
that NN O O
is NN O O
significantly NN O O
higher NN O O
than NN O O
that NN O O
of NN O O
the NN O O
baseline NN O I-Generic
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
that NN O O
sentence NN O I-Task
analysis NN O I-Task
should NN O O
be NN O O
treated NN O O
as NN O O
defeasible NN O I-Method
reasoning NN O I-Method
, NN O O
and NN O O
presents NN O O
such NN O O
a NN O O
treatment NN O I-Generic
for NN O O
Japanese NN O I-Task
sentence NN O I-Task
analyses NN O I-Task
using NN O O
an NN O O
argumentation NN O I-Method
system NN O I-Method
by NN O O
Konolige NN O O
, NN O O
which NN O O
is NN O O
a NN O O
formalization NN O I-Method
of NN O I-Method
defeasible NN O I-Method
reasoning NN O I-Method
, NN O O
that NN O O
includes NN O O
arguments NN O I-OtherScientificTerm
and NN O O
defeat NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
that NN O O
capture NN O O
defeasibility NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
research NN O O
on NN O O
the NN O O
Czech NN O I-Task
talking NN O I-Task
head NN O I-Task
system NN O I-Task
. NN O O

It NN O O
gives NN O O
an NN O O
overview NN O O
of NN O O
methods NN O I-Generic
used NN O O
for NN O O
visual NN O I-Task
speech NN O I-Task
animation NN O I-Task
, NN O O
parameterization NN O O
of NN O O
a NN O O
human NN O O
face NN O O
and NN O O
a NN O O
tongue NN O O
, NN O O
necessary NN O O
data NN O O
sources NN O O
and NN O O
a NN O O
synthesis NN O I-Method
method NN O I-Method
. NN O O

A NN O O
3D NN O I-Method
animation NN O I-Method
model NN O I-Method
is NN O O
used NN O O
for NN O O
a NN O O
pseudo-muscular NN O I-Method
animation NN O I-Method
schema NN O I-Method
to NN O O
create NN O O
such NN O O
animation NN O I-Task
of NN O I-Task
visual NN O I-Task
speech NN O I-Task
which NN O O
is NN O O
usable NN O O
for NN O O
a NN O O
lipreading NN O I-Task
. NN O O

An NN O O
extension NN O O
of NN O O
animation NN O I-Method
schema NN O I-Method
is NN O O
presented NN O O
to NN O O
reach NN O O
more NN O O
precise NN O O
deformations NN O O
mainly NN O O
in NN O O
a NN O O
lip NN O O
area NN O O
. NN O O

Furthermore NN O O
, NN O O
a NN O O
problem NN O O
of NN O O
forming NN O I-Task
articulatory NN O I-Task
trajectories NN O I-Task
is NN O O
formulated NN O O
to NN O O
solve NN O O
labial NN O I-OtherScientificTerm
coarticulation NN O I-OtherScientificTerm
effects NN O I-OtherScientificTerm
. NN O O

It NN O I-Generic
is NN O O
used NN O O
for NN O O
the NN O O
synthesis NN O I-Method
method NN O I-Method
based NN O O
on NN O O
a NN O O
selection NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
articulatory NN O I-OtherScientificTerm
targets NN O I-OtherScientificTerm
and NN O O
interpolation NN O I-Method
technique NN O I-Method
. NN O O

-DOCSTART- NN O O

It NN O O
is NN O O
often NN O O
assumed NN O O
that NN O O
when NN O O
natural NN O I-Task
language NN O I-Task
processing NN O I-Task
meets NN O O
the NN O O
real NN O O
world NN O O
, NN O O
the NN O O
ideal NN O O
of NN O O
aiming NN O O
for NN O O
complete NN O O
and NN O O
correct NN O O
interpretations NN O O
has NN O O
to NN O O
be NN O O
abandoned NN O O
. NN O O

However NN O O
, NN O O
our NN O O
experience NN O O
with NN O O
TACITUS NN O I-Method
; NN O O
especially NN O O
in NN O O
the NN O O
MUC-3 NN O I-Metric
evaluation NN O I-Metric
, NN O O
has NN O O
shown NN O O
that NN O O
principled NN O O
techniques NN O I-Generic
for NN O O
syntactic NN O I-Task
and NN O I-Task
pragmatic NN O I-Task
analysis NN O I-Task
can NN O O
be NN O O
bolstered NN O O
with NN O O
methods NN O I-Generic
for NN O O
achieving NN O O
robustness NN O I-Metric
. NN O O

We NN O O
describe NN O O
three NN O I-Generic
techniques NN O I-Generic
for NN O O
making NN O O
syntactic NN O I-Task
analysis NN O I-Task
more NN O O
robust NN O O
-- NN O O
an NN O O
agenda-based NN O I-Method
scheduling NN O I-Method
parser NN O I-Method
, NN O O
a NN O O
recovery NN O I-Method
technique NN O I-Method
for NN O O
failed NN O I-OtherScientificTerm
parses NN O I-OtherScientificTerm
, NN O O
and NN O O
a NN O O
new NN O O
technique NN O I-Generic
called NN O O
terminal NN O I-Method
substring NN O I-Method
parsing NN O I-Method
. NN O O

For NN O O
pragmatics NN O I-Task
processing NN O I-Task
, NN O O
we NN O O
describe NN O O
how NN O O
the NN O O
method NN O O
of NN O O
abductive NN O I-Method
inference NN O I-Method
is NN O O
inherently NN O O
robust NN O O
, NN O O
in NN O O
that NN O O
an NN O O
interpretation NN O O
is NN O O
always NN O O
possible NN O O
, NN O O
so NN O O
that NN O O
in NN O O
the NN O O
absence NN O O
of NN O O
the NN O O
required NN O O
world NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
, NN O O
performance NN O O
degrades NN O O
gracefully NN O O
. NN O O

Each NN O O
of NN O O
these NN O I-Generic
techniques NN O I-Generic
have NN O O
been NN O O
evaluated NN O O
and NN O O
the NN O O
results NN O O
of NN O O
the NN O O
evaluations NN O O
are NN O O
presented NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
a NN O O
Hidden NN O I-Method
Markov NN O I-Method
Model NN O I-Method
-LRB- NN O I-Method
HMM NN O I-Method
-RRB- NN O I-Method
and NN O O
an NN O O
HMM-based NN O I-Method
chunk NN O I-Method
tagger NN O I-Method
, NN O O
from NN O O
which NN O O
a NN O O
named NN O I-Method
entity NN O I-Method
-LRB- NN O I-Method
NE NN O I-Method
-RRB- NN O I-Method
recognition NN O I-Method
-LRB- NN O I-Method
NER NN O I-Method
-RRB- NN O I-Method
system NN O I-Method
is NN O O
built NN O O
to NN O O
recognize NN O O
and NN O O
classify NN O O
names NN O I-OtherScientificTerm
, NN O O
times NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
numerical NN O I-OtherScientificTerm
quantities NN O I-OtherScientificTerm
. NN O O

Through NN O O
the NN O O
HMM NN O I-Method
, NN O O
our NN O O
system NN O O
is NN O O
able NN O O
to NN O O
apply NN O O
and NN O O
integrate NN O O
four NN O O
types NN O O
of NN O O
internal NN O O
and NN O O
external NN O O
evidences NN O O
: NN O O
1 NN O O
-RRB- NN O O
simple NN O O
deterministic NN O I-OtherScientificTerm
internal NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
, NN O O
such NN O O
as NN O O
capitalization NN O I-OtherScientificTerm
and NN O O
digitalization NN O I-OtherScientificTerm
; NN O O
2 NN O O
-RRB- NN O O
internal NN O O
semantic NN O O
feature NN O O
of NN O O
important NN O O
triggers NN O O
; NN O O
3 NN O O
-RRB- NN O O
internal NN O I-OtherScientificTerm
gazetteer NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
; NN O O
4 NN O O
-RRB- NN O O
external NN O I-OtherScientificTerm
macro NN O I-OtherScientificTerm
context NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
way NN O O
, NN O O
the NN O O
NER NN O I-Task
problem NN O I-Task
can NN O O
be NN O O
resolved NN O O
effectively NN O O
. NN O O

Evaluation NN O O
of NN O O
our NN O O
system NN O I-Generic
on NN O O
MUC-6 NN O I-Material
and NN O I-Material
MUC-7 NN O I-Material
English NN O I-Material
NE NN O I-Material
tasks NN O I-Material
achieves NN O O
F-measures NN O I-Metric
of NN O O
96.6 NN O O
% NN O O
and NN O O
94.1 NN O O
% NN O O
respectively NN O O
. NN O O

It NN O O
shows NN O O
that NN O O
the NN O O
performance NN O O
is NN O O
significantly NN O O
better NN O O
than NN O O
reported NN O O
by NN O O
any NN O O
other NN O O
machine-learning NN O I-Method
system NN O I-Method
. NN O O

Moreover NN O O
, NN O O
the NN O O
performance NN O O
is NN O O
even NN O O
consistently NN O O
better NN O O
than NN O O
those NN O O
based NN O O
on NN O O
handcrafted NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Two NN O O
themes NN O I-Generic
have NN O O
evolved NN O O
in NN O O
speech NN O I-Task
and NN O I-Task
text NN O I-Task
image NN O I-Task
processing NN O I-Task
work NN O O
at NN O O
Xerox NN O O
PARC NN O O
that NN O O
expand NN O O
and NN O O
redefine NN O O
the NN O O
role NN O O
of NN O O
recognition NN O I-Method
technology NN O I-Method
in NN O O
document-oriented NN O I-Task
applications NN O I-Task
. NN O O

One NN O I-Generic
is NN O O
the NN O O
development NN O O
of NN O O
systems NN O I-Generic
that NN O O
provide NN O O
functionality NN O O
similar NN O O
to NN O O
that NN O O
of NN O O
text NN O I-Method
processors NN O I-Method
but NN O O
operate NN O O
directly NN O O
on NN O O
audio NN O I-Material
and NN O I-Material
scanned NN O I-Material
image NN O I-Material
data NN O I-Material
. NN O O

A NN O O
second NN O O
, NN O O
related NN O O
theme NN O I-Generic
is NN O O
the NN O O
use NN O O
of NN O O
speech NN O I-Method
and NN O I-Method
text-image NN O I-Method
recognition NN O I-Method
to NN O O
retrieve NN O O
arbitrary NN O O
, NN O O
user-specified NN O O
information NN O O
from NN O O
documents NN O I-OtherScientificTerm
with NN O I-OtherScientificTerm
signal NN O I-OtherScientificTerm
content NN O I-OtherScientificTerm
. NN O O

This NN O O
paper NN O O
discusses NN O O
three NN O O
research NN O I-Generic
initiatives NN O O
at NN O O
PARC NN O O
that NN O O
exemplify NN O O
these NN O O
themes NN O I-Generic
: NN O O
a NN O O
text-image NN O I-Method
editor NN O I-Method
-LSB- NN O O
1 NN O O
-RSB- NN O O
, NN O O
a NN O O
wordspotter NN O I-Method
for NN O O
voice NN O I-Task
editing NN O I-Task
and NN O I-Task
indexing NN O I-Task
-LSB- NN O O
12 NN O O
-RSB- NN O O
, NN O O
and NN O O
a NN O O
decoding NN O I-Method
framework NN O I-Method
for NN O O
scanned-document NN O I-Task
content NN O I-Task
retrieval NN O I-Task
-LSB- NN O O
4 NN O O
-RSB- NN O O
. NN O O

The NN O O
discussion NN O O
focuses NN O O
on NN O O
key NN O O
concepts NN O O
embodied NN O O
in NN O O
the NN O O
research NN O O
that NN O O
enable NN O O
novel NN O O
signal-based NN O I-Task
document NN O I-Task
processing NN O I-Task
functionality NN O I-Task
. NN O O

-DOCSTART- NN O O

The NN O O
problem NN O O
of NN O O
predicting NN O I-Task
image NN O I-Task
or NN O I-Task
video NN O I-Task
interestingness NN O I-Task
from NN O O
their NN O O
low-level NN O I-Method
feature NN O I-Method
representations NN O I-Method
has NN O O
received NN O O
increasing NN O O
interest NN O O
. NN O O

As NN O O
a NN O O
highly NN O O
subjective NN O I-OtherScientificTerm
visual NN O I-OtherScientificTerm
attribute NN O I-OtherScientificTerm
, NN O O
annotating NN O O
the NN O O
interesting-ness NN O I-OtherScientificTerm
value NN O I-OtherScientificTerm
of NN O O
training NN O O
data NN O O
for NN O O
learning NN O O
a NN O O
prediction NN O I-Method
model NN O I-Method
is NN O O
challenging NN O O
. NN O O

To NN O O
make NN O O
the NN O O
annotation NN O O
less NN O O
subjective NN O O
and NN O O
more NN O O
reliable NN O O
, NN O O
recent NN O O
studies NN O O
employ NN O O
crowdsourcing NN O I-Method
tools NN O I-Method
to NN O O
collect NN O O
pairwise NN O I-OtherScientificTerm
comparisons NN O I-OtherScientificTerm
-- NN O O
relying NN O O
on NN O O
majority NN O I-Method
voting NN O I-Method
to NN O O
prune NN O O
the NN O O
annotation NN O I-OtherScientificTerm
outliers/errors NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
more NN O O
principled NN O O
way NN O I-Generic
to NN O O
identify NN O O
annotation NN O I-OtherScientificTerm
outliers NN O I-OtherScientificTerm
by NN O O
formulating NN O O
the NN O O
interestingness NN O I-Task
prediction NN O I-Task
task NN O I-Task
as NN O O
a NN O O
unified NN O I-Method
robust NN O I-Method
learning NN O I-Method
to NN O O
rank NN O I-Task
problem NN O I-Task
, NN O O
tackling NN O O
both NN O O
the NN O O
outlier NN O I-Task
detection NN O I-Task
and NN O O
interestingness NN O I-Task
prediction NN O I-Task
tasks NN O I-Task
jointly NN O O
. NN O O

Extensive NN O O
experiments NN O O
on NN O O
both NN O O
image NN O I-Material
and NN O I-Material
video NN O I-Material
interestingness NN O I-Material
benchmark NN O I-Material
datasets NN O I-Material
demonstrate NN O O
that NN O O
our NN O O
new NN O O
approach NN O I-Generic
significantly NN O O
outperforms NN O O
state-of-the-art NN O I-Generic
alternatives NN O I-Generic
. NN O O

-DOCSTART- NN O O

Many NN O O
description NN O I-Method
logics NN O I-Method
-LRB- NN O I-Method
DLs NN O I-Method
-RRB- NN O I-Method
combine NN O O
knowledge NN O I-Method
representation NN O I-Method
on NN O O
an NN O O
abstract NN O O
, NN O O
logical NN O O
level NN O O
with NN O O
an NN O O
interface NN O O
to NN O O
`` NN O O
concrete NN O O
'' NN O O
domains NN O O
such NN O O
as NN O O
numbers NN O O
and NN O O
strings NN O O
. NN O O

We NN O O
propose NN O O
to NN O O
extend NN O O
such NN O O
DLs NN O I-Method
with NN O O
key NN O O
constraints NN O O
that NN O O
allow NN O O
the NN O O
expression NN O O
of NN O O
statements NN O O
like NN O O
`` NN O O
US NN O O
citizens NN O O
are NN O O
uniquely NN O O
identified NN O O
by NN O O
their NN O O
social NN O O
security NN O O
number NN O O
'' NN O O
. NN O O

Based NN O O
on NN O O
this NN O O
idea NN O O
, NN O O
we NN O O
introduce NN O O
a NN O O
number NN O O
of NN O O
natural NN O I-Method
description NN O I-Method
logics NN O I-Method
and NN O O
present NN O O
-LRB- NN O O
un NN O O
-RRB- NN O O
decidability NN O O
results NN O O
and NN O O
tight NN O I-OtherScientificTerm
NEx-PTlME NN O I-OtherScientificTerm
complexity NN O I-OtherScientificTerm
bounds NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
an NN O O
implementation NN O O
of NN O O
data-driven NN O I-Task
selection NN O I-Task
of NN O O
emphatic NN O I-Material
facial NN O I-Material
displays NN O I-Material
for NN O O
an NN O O
embodied NN O I-Task
conversational NN O I-Task
agent NN O I-Task
in NN O O
a NN O O
dialogue NN O I-Task
system NN O I-Task
. NN O O

A NN O O
corpus NN O I-Material
of NN O I-Material
sentences NN O I-Material
in NN O O
the NN O O
domain NN O O
of NN O O
the NN O O
target NN O O
dialogue NN O I-Method
system NN O I-Method
was NN O O
recorded NN O O
, NN O O
and NN O O
the NN O O
facial NN O I-Material
displays NN O I-Material
used NN O O
by NN O O
the NN O O
speaker NN O O
were NN O O
annotated NN O O
. NN O O

The NN O O
data NN O I-Generic
from NN O O
those NN O O
recordings NN O O
was NN O O
used NN O O
in NN O O
a NN O O
range NN O O
of NN O O
models NN O I-Generic
for NN O O
generating NN O O
facial NN O I-Material
displays NN O I-Material
, NN O O
each NN O O
model NN O I-Generic
making NN O O
use NN O O
of NN O O
a NN O O
different NN O O
amount NN O O
of NN O O
context NN O O
or NN O O
choosing NN O O
displays NN O O
differently NN O O
within NN O O
a NN O O
context NN O O
. NN O O

The NN O O
models NN O I-Generic
were NN O O
evaluated NN O O
in NN O O
two NN O O
ways NN O O
: NN O O
by NN O O
cross-validation NN O I-Method
against NN O O
the NN O O
corpus NN O I-Generic
, NN O O
and NN O O
by NN O O
asking NN O O
users NN O O
to NN O O
rate NN O O
the NN O O
output NN O O
. NN O O

The NN O O
predictions NN O O
of NN O O
the NN O O
cross-validation NN O I-Method
study NN O O
differed NN O O
from NN O O
the NN O O
actual NN O O
user NN O O
ratings NN O O
. NN O O

While NN O O
the NN O O
cross-validation NN O I-Method
gave NN O O
the NN O O
highest NN O O
scores NN O O
to NN O O
models NN O O
making NN O O
a NN O O
majority NN O O
choice NN O O
within NN O O
a NN O O
context NN O O
, NN O O
the NN O O
user NN O O
study NN O O
showed NN O O
a NN O O
significant NN O O
preference NN O O
for NN O O
models NN O O
that NN O O
produced NN O O
more NN O O
variation NN O O
. NN O O

This NN O O
preference NN O O
was NN O O
especially NN O O
strong NN O O
among NN O O
the NN O O
female NN O O
subjects NN O O
. NN O O

-DOCSTART- NN O O

When NN O O
classifying NN O I-Task
high-dimensional NN O I-Task
sequence NN O I-Task
data NN O I-Task
, NN O O
traditional NN O O
methods NN O O
-LRB- NN O O
e.g. NN O O
, NN O O
HMMs NN O I-Method
, NN O O
CRFs NN O I-Method
-RRB- NN O O
may NN O O
require NN O O
large NN O O
amounts NN O O
of NN O O
training NN O O
data NN O O
to NN O O
avoid NN O O
overfitting NN O I-OtherScientificTerm
. NN O O

In NN O O
such NN O O
cases NN O O
dimensionality NN O I-Method
reduction NN O I-Method
can NN O O
be NN O O
employed NN O O
to NN O O
find NN O O
a NN O O
low-dimensional NN O I-Method
representation NN O I-Method
on NN O O
which NN O O
classification NN O I-Task
can NN O O
be NN O O
done NN O O
more NN O O
efficiently NN O O
. NN O O

Existing NN O I-Generic
methods NN O I-Generic
for NN O O
supervised NN O I-Task
dimensionality NN O I-Task
reduction NN O I-Task
often NN O O
presume NN O O
that NN O O
the NN O O
data NN O O
is NN O O
densely NN O O
sampled NN O O
so NN O O
that NN O O
a NN O O
neighborhood NN O I-OtherScientificTerm
graph NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
can NN O O
be NN O O
formed NN O O
, NN O O
or NN O O
that NN O O
the NN O O
data NN O O
arises NN O O
from NN O O
a NN O O
known NN O I-OtherScientificTerm
distribution NN O I-OtherScientificTerm
. NN O O

Sufficient NN O I-Method
dimension NN O I-Method
reduction NN O I-Method
techniques NN O I-Method
aim NN O O
to NN O O
find NN O O
a NN O O
low NN O I-Method
dimensional NN O I-Method
representation NN O I-Method
such NN O O
that NN O O
the NN O O
remaining NN O O
degrees NN O O
of NN O O
freedom NN O O
become NN O O
conditionally NN O O
independent NN O O
of NN O O
the NN O O
output NN O O
values NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
develop NN O O
a NN O O
novel NN O O
sequence NN O I-Method
kernel NN O I-Method
dimension NN O I-Method
reduction NN O I-Method
approach NN O I-Method
-LRB- NN O I-Method
S-KDR NN O I-Method
-RRB- NN O I-Method
. NN O O

Our NN O O
approach NN O I-Generic
does NN O O
not NN O O
make NN O O
strong NN O O
assumptions NN O O
on NN O O
the NN O O
distribution NN O O
of NN O O
the NN O O
input NN O O
data NN O O
. NN O O

Spatial NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
temporal NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
periodic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
is NN O O
combined NN O O
in NN O O
a NN O O
principled NN O O
manner NN O O
, NN O O
and NN O O
an NN O O
optimal NN O O
manifold NN O I-OtherScientificTerm
is NN O O
learned NN O O
for NN O O
the NN O O
end-task NN O I-Generic
. NN O O

We NN O O
demonstrate NN O O
the NN O O
effectiveness NN O O
of NN O O
our NN O O
approach NN O I-Generic
on NN O O
several NN O O
tasks NN O O
involving NN O O
the NN O O
discrimination NN O I-Task
of NN O I-Task
human NN O I-Task
gesture NN O I-Task
and NN O I-Task
motion NN O I-Task
categories NN O I-Task
, NN O O
as NN O O
well NN O O
as NN O O
on NN O O
a NN O O
database NN O I-Material
of NN O I-Material
dynamic NN O I-Material
textures NN O I-Material
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
an NN O O
efficient NN O O
algorithm NN O I-Generic
for NN O O
chart-based NN O I-Task
phrase NN O I-Task
structure NN O I-Task
parsing NN O I-Task
of NN O O
natural NN O I-Material
language NN O I-Material
that NN O O
is NN O O
tailored NN O O
to NN O O
the NN O O
problem NN O O
of NN O O
extracting NN O O
specific NN O O
information NN O O
from NN O O
unrestricted NN O O
texts NN O O
where NN O O
many NN O O
of NN O O
the NN O O
words NN O O
are NN O O
unknown NN O O
and NN O O
much NN O O
of NN O O
the NN O O
text NN O O
is NN O O
irrelevant NN O O
to NN O O
the NN O O
task NN O O
. NN O O

The NN O O
parser NN O I-Method
gains NN O O
algorithmic NN O O
efficiency NN O O
through NN O O
a NN O O
reduction NN O O
of NN O O
its NN O O
search NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
. NN O O

As NN O O
each NN O O
new NN O O
edge NN O I-OtherScientificTerm
is NN O O
added NN O O
to NN O O
the NN O O
chart NN O I-OtherScientificTerm
, NN O O
the NN O O
algorithm NN O O
checks NN O O
only NN O O
the NN O O
topmost NN O O
of NN O O
the NN O O
edges NN O I-OtherScientificTerm
adjacent NN O O
to NN O O
it NN O O
, NN O O
rather NN O O
than NN O O
all NN O O
such NN O O
edges NN O I-OtherScientificTerm
as NN O O
in NN O O
conventional NN O O
treatments NN O O
. NN O O

The NN O O
resulting NN O O
spanning NN O I-OtherScientificTerm
edges NN O I-OtherScientificTerm
are NN O O
insured NN O O
to NN O O
be NN O O
the NN O O
correct NN O O
ones NN O O
by NN O O
carefully NN O O
controlling NN O O
the NN O O
order NN O O
in NN O O
which NN O O
edges NN O I-OtherScientificTerm
are NN O O
introduced NN O O
so NN O O
that NN O O
every NN O O
final NN O O
constituent NN O O
covers NN O O
the NN O O
longest NN O O
possible NN O O
span NN O O
. NN O O

This NN O O
is NN O O
facilitated NN O O
through NN O O
the NN O O
use NN O O
of NN O O
phrase NN O I-Method
boundary NN O I-Method
heuristics NN O I-Method
based NN O O
on NN O O
the NN O O
placement NN O O
of NN O O
function NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
, NN O O
and NN O O
by NN O O
heuristic NN O I-Method
rules NN O I-Method
that NN O O
permit NN O O
certain NN O O
kinds NN O O
of NN O O
phrases NN O O
to NN O O
be NN O O
deduced NN O O
despite NN O O
the NN O O
presence NN O O
of NN O O
unknown NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
. NN O O

A NN O O
further NN O O
reduction NN O I-OtherScientificTerm
in NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
search NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
is NN O O
achieved NN O O
by NN O O
using NN O O
semantic NN O I-OtherScientificTerm
rather NN O O
than NN O O
syntactic NN O I-OtherScientificTerm
categories NN O I-OtherScientificTerm
on NN O O
the NN O O
terminal NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
non-terminal NN O I-OtherScientificTerm
edges NN O I-OtherScientificTerm
, NN O O
thereby NN O O
reducing NN O O
the NN O O
amount NN O O
of NN O O
ambiguity NN O O
and NN O O
thus NN O O
the NN O O
number NN O O
of NN O O
edges NN O I-OtherScientificTerm
, NN O O
since NN O O
only NN O O
edges NN O I-OtherScientificTerm
with NN O O
a NN O O
valid NN O O
semantic NN O O
interpretation NN O O
are NN O O
ever NN O O
introduced NN O O
. NN O O

-DOCSTART- NN O O

Automatic NN O I-Task
estimation NN O I-Task
of NN O I-Task
word NN O I-OtherScientificTerm
significance NN O I-OtherScientificTerm
oriented NN O O
for NN O O
speech-based NN O I-Task
Information NN O I-Task
Retrieval NN O I-Task
-LRB- NN O I-Task
IR NN O I-Task
-RRB- NN O I-Task
is NN O O
addressed NN O O
. NN O O

Since NN O O
the NN O O
significance NN O I-OtherScientificTerm
of NN O O
words NN O O
differs NN O O
in NN O O
IR NN O I-Task
, NN O O
automatic NN O I-Task
speech NN O I-Task
recognition NN O I-Task
-LRB- NN O I-Task
ASR NN O I-Task
-RRB- NN O I-Task
performance NN O O
has NN O O
been NN O O
evaluated NN O O
based NN O O
on NN O O
weighted NN O I-Metric
word NN O I-Metric
error NN O I-Metric
rate NN O I-Metric
-LRB- NN O I-Metric
WWER NN O I-Metric
-RRB- NN O I-Metric
, NN O O
which NN O O
gives NN O O
a NN O O
weight NN O O
on NN O O
errors NN O O
from NN O O
the NN O O
viewpoint NN O O
of NN O O
IR NN O I-Task
, NN O O
instead NN O O
of NN O O
word NN O I-Metric
error NN O I-Metric
rate NN O I-Metric
-LRB- NN O I-Metric
WER NN O I-Metric
-RRB- NN O I-Metric
, NN O O
which NN O O
treats NN O O
all NN O O
words NN O O
uniformly NN O O
. NN O O

A NN O O
decoding NN O I-Method
strategy NN O I-Method
that NN O O
minimizes NN O O
WWER NN O I-Metric
based NN O O
on NN O O
a NN O O
Minimum NN O I-Method
Bayes-Risk NN O I-Method
framework NN O I-Method
has NN O O
been NN O O
shown NN O O
, NN O O
and NN O O
the NN O O
reduction NN O O
of NN O O
errors NN O O
on NN O O
both NN O O
ASR NN O I-Task
and NN O O
IR NN O I-Task
has NN O O
been NN O O
reported NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
an NN O O
automatic NN O I-Method
estimation NN O I-Method
method NN O I-Method
for NN O O
word NN O I-OtherScientificTerm
significance NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
weights NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
based NN O O
on NN O O
its NN O O
influence NN O O
on NN O O
IR NN O I-Task
. NN O O

Specifically NN O O
, NN O O
weights NN O O
are NN O O
estimated NN O O
so NN O O
that NN O O
evaluation NN O I-Metric
measures NN O I-Metric
of NN O O
ASR NN O I-Task
and NN O O
IR NN O I-Task
are NN O O
equivalent NN O O
. NN O O

We NN O O
apply NN O O
the NN O O
proposed NN O O
method NN O I-Generic
to NN O O
a NN O O
speech-based NN O I-Task
information NN O I-Task
retrieval NN O I-Task
system NN O I-Task
, NN O O
which NN O O
is NN O O
a NN O O
typical NN O O
IR NN O I-Task
system NN O I-Task
, NN O O
and NN O O
show NN O O
that NN O O
the NN O O
method NN O I-Generic
works NN O O
well NN O O
. NN O O

-DOCSTART- NN O O

Methods NN O I-Generic
developed NN O O
for NN O O
spelling NN O I-Task
correction NN O I-Task
for NN O O
languages NN O I-OtherScientificTerm
like NN O O
English NN O I-OtherScientificTerm
-LRB- NN O O
see NN O O
the NN O O
review NN O O
by NN O O
Kukich NN O O
-LRB- NN O O
Kukich NN O O
, NN O O
1992 NN O O
-RRB- NN O O
-RRB- NN O O
are NN O O
not NN O O
readily NN O O
applicable NN O O
to NN O O
agglutinative NN O I-OtherScientificTerm
languages NN O I-OtherScientificTerm
. NN O O

This NN O O
poster NN O O
presents NN O O
an NN O O
approach NN O I-Generic
to NN O O
spelling NN O I-Task
correction NN O I-Task
in NN O O
agglutinative NN O I-OtherScientificTerm
languages NN O I-OtherScientificTerm
that NN O O
is NN O O
based NN O O
on NN O O
two-level NN O I-OtherScientificTerm
morphology NN O I-OtherScientificTerm
and NN O O
a NN O O
dynamic-programming NN O I-Method
based NN O I-Method
search NN O I-Method
algorithm NN O I-Method
. NN O O

After NN O O
an NN O O
overview NN O O
of NN O O
our NN O O
approach NN O I-Generic
, NN O O
we NN O O
present NN O O
results NN O O
from NN O O
experiments NN O O
with NN O O
spelling NN O I-Task
correction NN O I-Task
in NN O O
Turkish NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
a NN O O
novel NN O O
training NN O I-Method
method NN O I-Method
for NN O O
a NN O O
localized NN O I-Method
phrase-based NN O I-Method
prediction NN O I-Method
model NN O I-Method
for NN O O
statistical NN O I-Task
machine NN O I-Task
translation NN O I-Task
-LRB- NN O I-Task
SMT NN O I-Task
-RRB- NN O I-Task
. NN O O

The NN O O
model NN O I-Generic
predicts NN O O
blocks NN O O
with NN O O
orientation NN O O
to NN O O
handle NN O O
local NN O I-Task
phrase NN O I-Task
re-ordering NN O I-Task
. NN O O

We NN O O
use NN O O
a NN O O
maximum NN O I-OtherScientificTerm
likelihood NN O I-OtherScientificTerm
criterion NN O I-OtherScientificTerm
to NN O O
train NN O O
a NN O O
log-linear NN O I-Method
block NN O I-Method
bigram NN O I-Method
model NN O I-Method
which NN O O
uses NN O O
real-valued NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
-LRB- NN O O
e.g. NN O O
a NN O O
language NN O I-OtherScientificTerm
model NN O I-OtherScientificTerm
score NN O I-OtherScientificTerm
-RRB- NN O O
as NN O O
well NN O O
as NN O O
binary NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
based NN O O
on NN O O
the NN O O
block NN O O
identities NN O O
themselves NN O O
, NN O O
e.g. NN O O
block NN O O
bigram NN O O
features NN O O
. NN O O

Our NN O O
training NN O I-Method
algorithm NN O I-Method
can NN O O
easily NN O O
handle NN O O
millions NN O O
of NN O O
features NN O I-OtherScientificTerm
. NN O O

The NN O O
best NN O O
system NN O I-Generic
obtains NN O O
a NN O O
18.6 NN O O
% NN O O
improvement NN O O
over NN O O
the NN O O
baseline NN O I-Generic
on NN O O
a NN O O
standard NN O O
Arabic-English NN O I-Task
translation NN O I-Task
task NN O I-Task
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
describe NN O O
a NN O O
novel NN O O
data NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
for NN O O
phrase-based NN O I-Task
statistical NN O I-Task
machine NN O I-Task
translation NN O I-Task
which NN O O
allows NN O O
for NN O O
the NN O O
retrieval NN O I-Task
of NN O I-Task
arbitrarily NN O I-Task
long NN O I-Task
phrases NN O I-Task
while NN O O
simultaneously NN O O
using NN O O
less NN O O
memory NN O I-OtherScientificTerm
than NN O O
is NN O O
required NN O O
by NN O O
current NN O O
decoder NN O I-Method
implementations NN O O
. NN O O

We NN O O
detail NN O O
the NN O O
computational NN O I-Metric
complexity NN O I-Metric
and NN O O
average NN O I-Metric
retrieval NN O I-Metric
times NN O I-Metric
for NN O O
looking NN O O
up NN O O
phrase NN O I-OtherScientificTerm
translations NN O I-OtherScientificTerm
in NN O O
our NN O O
suffix NN O I-OtherScientificTerm
array-based NN O I-OtherScientificTerm
data NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
. NN O O

We NN O O
show NN O O
how NN O O
sampling NN O I-Method
can NN O O
be NN O O
used NN O O
to NN O O
reduce NN O O
the NN O O
retrieval NN O I-Metric
time NN O I-Metric
by NN O O
orders NN O O
of NN O O
magnitude NN O O
with NN O O
no NN O O
loss NN O O
in NN O O
translation NN O I-Metric
quality NN O I-Metric
. NN O O

-DOCSTART- NN O O

The NN O O
major NN O O
objective NN O O
of NN O O
this NN O O
program NN O O
is NN O O
to NN O O
develop NN O O
and NN O O
demonstrate NN O O
robust NN O O
, NN O O
high NN O O
performance NN O O
continuous NN O I-Method
speech NN O I-Method
recognition NN O I-Method
-LRB- NN O I-Method
CSR NN O I-Method
-RRB- NN O I-Method
techniques NN O I-Method
focussed NN O O
on NN O O
application NN O O
in NN O O
Spoken NN O I-Task
Language NN O I-Task
Systems NN O I-Task
-LRB- NN O I-Task
SLS NN O I-Task
-RRB- NN O I-Task
which NN O O
will NN O O
enhance NN O O
the NN O O
effectiveness NN O O
of NN O O
military NN O I-Task
and NN O I-Task
civilian NN O I-Task
computer-based NN O I-Task
systems NN O I-Task
. NN O O

A NN O O
key NN O O
complementary NN O O
objective NN O O
is NN O O
to NN O O
define NN O O
and NN O O
develop NN O O
applications NN O O
of NN O O
robust NN O O
speech NN O I-Task
recognition NN O I-Task
and NN O I-Task
understanding NN O I-Task
systems NN O I-Task
, NN O O
and NN O O
to NN O O
help NN O O
catalyze NN O O
the NN O O
transition NN O O
of NN O O
spoken NN O I-Method
language NN O I-Method
technology NN O I-Method
into NN O O
military NN O I-Task
and NN O I-Task
civilian NN O I-Task
systems NN O I-Task
, NN O O
with NN O O
particular NN O O
focus NN O O
on NN O O
application NN O O
of NN O O
robust NN O O
CSR NN O I-Method
to NN O O
mobile NN O I-Task
military NN O I-Task
command NN O I-Task
and NN O I-Task
control NN O I-Task
. NN O O

The NN O O
research NN O O
effort NN O O
focusses NN O O
on NN O O
developing NN O O
advanced NN O O
acoustic NN O I-Method
modelling NN O I-Method
, NN O O
rapid NN O I-Method
search NN O I-Method
, NN O O
and NN O O
recognition-time NN O I-Method
adaptation NN O I-Method
techniques NN O I-Method
for NN O O
robust NN O O
large-vocabulary NN O I-Method
CSR NN O I-Method
, NN O O
and NN O O
on NN O O
applying NN O O
these NN O O
techniques NN O I-Generic
to NN O O
the NN O O
new NN O O
ARPA NN O I-Material
large-vocabulary NN O I-Material
CSR NN O I-Material
corpora NN O I-Material
and NN O O
to NN O O
military NN O I-Task
application NN O I-Task
tasks NN O I-Task
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
examines NN O O
what NN O O
kind NN O O
of NN O O
similarity NN O I-OtherScientificTerm
between NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
can NN O O
be NN O O
represented NN O O
by NN O O
what NN O O
kind NN O O
of NN O O
word NN O I-OtherScientificTerm
vectors NN O I-OtherScientificTerm
in NN O O
the NN O O
vector NN O I-Method
space NN O I-Method
model NN O I-Method
. NN O O

Through NN O O
two NN O O
experiments NN O O
, NN O O
three NN O O
methods NN O I-Generic
for NN O O
constructing NN O I-Task
word NN O I-Task
vectors NN O I-Task
, NN O O
i.e. NN O O
, NN O O
LSA-based NN O I-Method
, NN O I-Method
cooccurrence-based NN O I-Method
and NN O I-Method
dictionary-based NN O I-Method
methods NN O I-Method
, NN O O
were NN O O
compared NN O O
in NN O O
terms NN O O
of NN O O
the NN O O
ability NN O O
to NN O O
represent NN O O
two NN O O
kinds NN O O
of NN O O
similarity NN O I-Generic
, NN O O
i.e. NN O O
, NN O O
taxonomic NN O I-OtherScientificTerm
similarity NN O I-OtherScientificTerm
and NN O O
associative NN O I-OtherScientificTerm
similarity NN O I-OtherScientificTerm
. NN O O

The NN O O
result NN O O
of NN O O
the NN O O
comparison NN O O
was NN O O
that NN O O
the NN O O
dictionary-based NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
vectors NN O I-OtherScientificTerm
better NN O O
reflect NN O O
taxonomic NN O I-OtherScientificTerm
similarity NN O I-OtherScientificTerm
, NN O O
while NN O O
the NN O O
LSA-based NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
cooccurrence-based NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
vectors NN O I-OtherScientificTerm
better NN O O
reflect NN O O
associative NN O I-OtherScientificTerm
similarity NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
maximum NN O I-Method
entropy NN O I-Method
word NN O I-Method
alignment NN O I-Method
algorithm NN O I-Method
for NN O O
Arabic-English NN O I-Material
based NN O O
on NN O O
supervised NN O I-Material
training NN O I-Material
data NN O I-Material
. NN O O

We NN O O
demonstrate NN O O
that NN O O
it NN O O
is NN O O
feasible NN O O
to NN O O
create NN O O
training NN O I-Material
material NN O I-Material
for NN O O
problems NN O O
in NN O O
machine NN O I-Task
translation NN O I-Task
and NN O O
that NN O O
a NN O O
mixture NN O O
of NN O O
supervised NN O I-Method
and NN O I-Method
unsupervised NN O I-Method
methods NN O I-Method
yields NN O O
superior NN O O
performance NN O O
. NN O O

The NN O O
probabilistic NN O I-Method
model NN O I-Method
used NN O O
in NN O O
the NN O O
alignment NN O I-Task
directly NN O O
models NN O O
the NN O O
link NN O I-Task
decisions NN O I-Task
. NN O O

Significant NN O O
improvement NN O O
over NN O O
traditional NN O O
word NN O I-Method
alignment NN O I-Method
techniques NN O I-Method
is NN O O
shown NN O O
as NN O O
well NN O O
as NN O O
improvement NN O O
on NN O O
several NN O O
machine NN O I-OtherScientificTerm
translation NN O I-OtherScientificTerm
tests NN O I-OtherScientificTerm
. NN O O

Performance NN O O
of NN O O
the NN O O
algorithm NN O I-Generic
is NN O O
contrasted NN O O
with NN O O
human NN O I-OtherScientificTerm
annotation NN O I-OtherScientificTerm
performance NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
novel NN O O
Cooperative NN O I-Method
Model NN O I-Method
for NN O O
natural NN O I-Task
language NN O I-Task
understanding NN O I-Task
in NN O O
a NN O O
dialogue NN O I-Task
system NN O I-Task
. NN O O

We NN O O
build NN O O
this NN O I-Generic
based NN O O
on NN O O
both NN O O
Finite NN O I-Method
State NN O I-Method
Model NN O I-Method
-LRB- NN O I-Method
FSM NN O I-Method
-RRB- NN O I-Method
and NN O O
Statistical NN O I-Method
Learning NN O I-Method
Model NN O I-Method
-LRB- NN O I-Method
SLM NN O I-Method
-RRB- NN O I-Method
. NN O O

FSM NN O I-Method
provides NN O O
two NN O O
strategies NN O O
for NN O O
language NN O I-Task
understanding NN O I-Task
and NN O O
have NN O O
a NN O O
high NN O O
accuracy NN O O
but NN O O
little NN O O
robustness NN O O
and NN O O
flexibility NN O O
. NN O O

Statistical NN O I-Method
approach NN O I-Method
is NN O O
much NN O O
more NN O O
robust NN O O
but NN O O
less NN O O
accurate NN O O
. NN O O

Cooperative NN O I-Method
Model NN O I-Method
incorporates NN O O
all NN O O
the NN O O
three NN O O
strategies NN O O
together NN O O
and NN O O
thus NN O O
can NN O O
suppress NN O O
all NN O O
the NN O O
shortcomings NN O O
of NN O O
different NN O O
strategies NN O O
and NN O O
has NN O O
all NN O O
the NN O O
advantages NN O O
of NN O O
the NN O O
three NN O O
strategies NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
the NN O O
Chinese NN O I-Material
language NN O I-Material
, NN O O
a NN O O
verb NN O O
may NN O O
have NN O O
its NN O O
dependents NN O O
on NN O O
its NN O O
left NN O O
, NN O O
right NN O O
or NN O O
on NN O O
both NN O O
sides NN O O
. NN O O

The NN O O
ambiguity NN O I-Task
resolution NN O I-Task
of NN O I-Task
right-side NN O I-Task
dependencies NN O I-Task
is NN O O
essential NN O O
for NN O O
dependency NN O I-Task
parsing NN O I-Task
of NN O O
sentences NN O O
with NN O O
two NN O O
or NN O O
more NN O O
verbs NN O O
. NN O O

Previous NN O O
works NN O O
on NN O O
shift-reduce NN O I-Method
dependency NN O I-Method
parsers NN O I-Method
may NN O O
not NN O O
guarantee NN O O
the NN O O
connectivity NN O I-Metric
of NN O O
a NN O O
dependency NN O I-OtherScientificTerm
tree NN O I-OtherScientificTerm
due NN O O
to NN O O
their NN O O
weakness NN O O
at NN O O
resolving NN O O
the NN O O
right-side NN O I-OtherScientificTerm
dependencies NN O I-OtherScientificTerm
. NN O O

This NN O O
paper NN O O
proposes NN O O
a NN O O
two-phase NN O I-Method
shift-reduce NN O I-Method
dependency NN O I-Method
parser NN O I-Method
based NN O O
on NN O O
SVM NN O I-Method
learning NN O I-Method
. NN O O

The NN O O
left-side NN O I-OtherScientificTerm
dependents NN O I-OtherScientificTerm
and NN O O
right-side NN O I-OtherScientificTerm
nominal NN O I-OtherScientificTerm
dependents NN O I-OtherScientificTerm
are NN O O
detected NN O O
in NN O O
Phase NN O O
I NN O O
, NN O O
and NN O O
right-side NN O I-OtherScientificTerm
verbal NN O I-OtherScientificTerm
dependents NN O I-OtherScientificTerm
are NN O O
decided NN O O
in NN O O
Phase NN O O
II NN O O
. NN O O

In NN O O
experimental NN O O
evaluation NN O O
, NN O O
our NN O O
proposed NN O O
method NN O I-Generic
outperforms NN O O
previous NN O O
shift-reduce NN O I-Method
dependency NN O I-Method
parsers NN O I-Method
for NN O O
the NN O O
Chine NN O I-Material
language NN O I-Material
, NN O O
showing NN O O
improvement NN O O
of NN O O
dependency NN O I-Metric
accuracy NN O I-Metric
by NN O O
10.08 NN O O
% NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
order NN O O
to NN O O
meet NN O O
the NN O O
needs NN O O
of NN O O
a NN O O
publication NN O O
of NN O O
papers NN O O
in NN O O
English NN O O
, NN O O
many NN O O
systems NN O O
to NN O O
run NN O O
off NN O O
texts NN O O
have NN O O
been NN O O
developed NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
report NN O O
a NN O O
system NN O I-Generic
FROFF NN O I-Method
which NN O O
can NN O O
make NN O O
a NN O O
fair NN O O
copy NN O O
of NN O O
not NN O O
only NN O O
texts NN O O
but NN O O
also NN O O
graphs NN O O
and NN O O
tables NN O O
indispensable NN O O
to NN O O
our NN O O
papers NN O O
. NN O O

Its NN O O
selection NN O O
of NN O O
fonts NN O O
, NN O O
specification NN O O
of NN O O
character NN O O
size NN O O
are NN O O
dynamically NN O O
changeable NN O O
, NN O O
and NN O O
the NN O O
typing NN O I-OtherScientificTerm
location NN O I-OtherScientificTerm
can NN O O
be NN O O
also NN O O
changed NN O O
in NN O O
lateral NN O O
or NN O O
longitudinal NN O O
directions NN O O
. NN O O

Each NN O O
character NN O O
has NN O O
its NN O O
own NN O O
width NN O O
and NN O O
a NN O O
line NN O O
length NN O O
is NN O O
counted NN O O
by NN O O
the NN O O
sum NN O O
of NN O O
each NN O O
character NN O O
. NN O O

By NN O O
using NN O O
commands NN O I-OtherScientificTerm
or NN O O
rules NN O I-OtherScientificTerm
which NN O O
are NN O O
defined NN O O
to NN O O
facilitate NN O O
the NN O O
construction NN O O
of NN O O
format NN O O
expected NN O O
or NN O O
some NN O O
mathematical NN O I-OtherScientificTerm
expressions NN O I-OtherScientificTerm
, NN O O
elaborate NN O O
and NN O O
pretty NN O O
documents NN O O
can NN O O
be NN O O
successfully NN O O
obtained NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
an NN O O
evaluation NN O I-Generic
method NN O I-Generic
employing NN O O
a NN O O
latent NN O I-Method
variable NN O I-Method
model NN O I-Method
for NN O O
paraphrases NN O I-OtherScientificTerm
with NN O O
their NN O O
contexts NN O O
. NN O O

We NN O O
assume NN O O
that NN O O
the NN O O
context NN O O
of NN O O
a NN O O
sentence NN O O
is NN O O
indicated NN O O
by NN O O
a NN O O
latent NN O I-OtherScientificTerm
variable NN O I-OtherScientificTerm
of NN O O
the NN O O
model NN O I-Generic
as NN O O
a NN O O
topic NN O O
and NN O O
that NN O O
the NN O O
likelihood NN O O
of NN O O
each NN O O
variable NN O O
can NN O O
be NN O O
inferred NN O O
. NN O O

A NN O O
paraphrase NN O I-OtherScientificTerm
is NN O O
evaluated NN O O
for NN O O
whether NN O O
its NN O O
sentences NN O O
are NN O O
used NN O O
in NN O O
the NN O O
same NN O O
context NN O O
. NN O O

Experimental NN O O
results NN O O
showed NN O O
that NN O O
the NN O O
proposed NN O O
method NN O I-Generic
achieves NN O O
almost NN O O
60 NN O O
% NN O O
accuracy NN O O
and NN O O
that NN O O
there NN O O
is NN O O
not NN O O
a NN O O
large NN O O
performance NN O O
difference NN O O
between NN O O
the NN O O
two NN O O
models NN O O
. NN O O

The NN O O
results NN O O
also NN O O
revealed NN O O
an NN O O
upper NN O O
bound NN O O
of NN O O
accuracy NN O I-Metric
of NN O O
77 NN O O
% NN O O
with NN O O
the NN O O
method NN O I-Generic
when NN O O
using NN O O
only NN O O
topic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
the NN O O
methods NN O I-Generic
and NN O O
hardware NN O I-Generic
that NN O O
we NN O O
are NN O O
using NN O O
to NN O O
produce NN O O
a NN O O
real-time NN O O
demonstration NN O O
of NN O O
an NN O O
integrated NN O I-Method
Spoken NN O I-Method
Language NN O I-Method
System NN O I-Method
. NN O O

We NN O O
describe NN O O
algorithms NN O I-Generic
that NN O O
greatly NN O O
reduce NN O O
the NN O O
computation NN O O
needed NN O O
to NN O O
compute NN O O
the NN O O
N-Best NN O I-OtherScientificTerm
sentence NN O I-OtherScientificTerm
hypotheses NN O I-OtherScientificTerm
. NN O O

To NN O O
avoid NN O O
grammar NN O I-OtherScientificTerm
coverage NN O I-OtherScientificTerm
problems NN O I-OtherScientificTerm
we NN O O
use NN O O
a NN O O
fully-connected NN O I-Method
first-order NN O I-Method
statistical NN O I-Method
class NN O I-Method
grammar NN O I-Method
. NN O O

The NN O O
speech-search NN O I-Method
algorithm NN O I-Method
is NN O O
implemented NN O O
on NN O O
a NN O O
board NN O I-OtherScientificTerm
with NN O O
a NN O O
single NN O O
Intel NN O I-OtherScientificTerm
i860 NN O I-OtherScientificTerm
chip NN O I-OtherScientificTerm
, NN O O
which NN O O
provides NN O O
a NN O O
factor NN O O
of NN O O
5 NN O O
speed-up NN O O
over NN O O
a NN O O
SUN NN O I-OtherScientificTerm
4 NN O I-OtherScientificTerm
for NN O O
straight NN O I-OtherScientificTerm
C NN O I-OtherScientificTerm
code NN O I-OtherScientificTerm
. NN O O

The NN O O
board NN O I-OtherScientificTerm
plugs NN O O
directly NN O O
into NN O O
the NN O O
VME NN O I-OtherScientificTerm
bus NN O I-OtherScientificTerm
of NN O O
the NN O O
SUN4 NN O I-OtherScientificTerm
, NN O O
which NN O O
controls NN O O
the NN O O
system NN O I-Generic
and NN O O
contains NN O O
the NN O O
natural NN O I-Method
language NN O I-Method
system NN O I-Method
and NN O O
application NN O I-OtherScientificTerm
back NN O I-OtherScientificTerm
end NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

We NN O O
address NN O O
the NN O O
problem NN O O
of NN O O
estimating NN O I-Task
location NN O I-Task
information NN O I-Task
of NN O O
an NN O O
image NN O I-Material
using NN O O
principles NN O O
from NN O O
automated NN O I-Method
representation NN O I-Method
learning NN O I-Method
. NN O O

We NN O O
pursue NN O O
a NN O O
hierarchical NN O I-Method
sparse NN O I-Method
coding NN O I-Method
approach NN O I-Method
that NN O O
learns NN O O
features NN O I-OtherScientificTerm
useful NN O O
in NN O O
discriminating NN O O
images NN O O
across NN O O
locations NN O O
, NN O O
by NN O O
initializing NN O O
it NN O I-Generic
with NN O O
a NN O O
geometric NN O I-OtherScientificTerm
prior NN O I-OtherScientificTerm
corresponding NN O O
to NN O O
transformations NN O O
between NN O O
image NN O I-OtherScientificTerm
appearance NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
and NN O O
their NN O O
corresponding NN O O
location NN O I-OtherScientificTerm
grouping NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
using NN O O
the NN O O
notion NN O O
of NN O O
parallel NN O I-OtherScientificTerm
transport NN O I-OtherScientificTerm
on NN O I-OtherScientificTerm
manifolds NN O I-OtherScientificTerm
. NN O O

We NN O O
then NN O O
extend NN O O
this NN O O
approach NN O I-Generic
to NN O O
account NN O O
for NN O O
the NN O O
availability NN O O
of NN O O
heterogeneous NN O I-OtherScientificTerm
data NN O I-OtherScientificTerm
modalities NN O I-OtherScientificTerm
such NN O O
as NN O O
geo-tags NN O I-OtherScientificTerm
and NN O O
videos NN O I-OtherScientificTerm
pertaining NN O O
to NN O O
different NN O O
locations NN O O
, NN O O
and NN O O
also NN O O
study NN O O
a NN O O
relatively NN O O
under-addressed NN O O
problem NN O O
of NN O O
transferring NN O I-Task
knowledge NN O I-Task
available NN O O
from NN O O
certain NN O O
locations NN O O
to NN O O
infer NN O O
the NN O O
grouping NN O I-Task
of NN O I-Task
data NN O I-Task
from NN O O
novel NN O O
locations NN O O
. NN O O

We NN O O
evaluate NN O O
our NN O O
approach NN O I-Method
on NN O O
several NN O O
standard NN O O
datasets NN O I-Generic
such NN O O
as NN O O
im2gps NN O I-Material
, NN O O
San NN O I-Material
Francisco NN O I-Material
and NN O O
MediaEval2010 NN O I-Material
, NN O O
and NN O O
obtain NN O O
state-of-the-art NN O O
results NN O O
. NN O O

-DOCSTART- NN O O

Conventional NN O O
HMMs NN O I-Method
have NN O O
weak NN O I-OtherScientificTerm
duration NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
. NN O O

In NN O O
noisy NN O O
conditions NN O O
, NN O O
the NN O O
mismatch NN O O
between NN O O
corrupted NN O I-Material
speech NN O I-Material
signals NN O I-Material
and NN O O
models NN O I-Generic
trained NN O O
on NN O O
clean NN O I-Material
speech NN O I-Material
may NN O O
cause NN O O
the NN O O
decoder NN O I-Method
to NN O O
produce NN O O
word NN O I-OtherScientificTerm
matches NN O I-OtherScientificTerm
with NN O O
unrealistic NN O I-OtherScientificTerm
durations NN O I-OtherScientificTerm
. NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
simple NN O O
way NN O O
to NN O O
incorporate NN O O
word NN O I-OtherScientificTerm
duration NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
by NN O O
unrolling NN O I-Method
HMMs NN O I-Method
to NN O O
form NN O O
a NN O O
lattice NN O I-OtherScientificTerm
where NN O O
word NN O I-OtherScientificTerm
duration NN O I-OtherScientificTerm
probabilities NN O I-OtherScientificTerm
can NN O O
be NN O O
applied NN O O
directly NN O O
to NN O O
state NN O I-OtherScientificTerm
transitions NN O I-OtherScientificTerm
. NN O O

The NN O O
expanded NN O O
HMMs NN O I-Method
are NN O O
compatible NN O O
with NN O O
conventional NN O O
Viterbi NN O I-Method
decoding NN O I-Method
. NN O O

Experiments NN O O
on NN O O
connected-digit NN O I-Task
recognition NN O I-Task
show NN O O
that NN O O
when NN O O
using NN O O
explicit NN O O
duration NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
the NN O O
decoder NN O I-Method
generates NN O O
word NN O I-OtherScientificTerm
matches NN O I-OtherScientificTerm
with NN O O
more NN O O
reasonable NN O O
durations NN O O
, NN O O
and NN O O
word NN O I-Metric
error NN O I-Metric
rates NN O I-Metric
are NN O O
significantly NN O O
reduced NN O O
across NN O O
a NN O O
broad NN O O
range NN O O
of NN O O
noise NN O I-OtherScientificTerm
conditions NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Soames NN O O
1979 NN O O
provides NN O O
some NN O O
counterexamples NN O O
to NN O O
the NN O O
theory NN O O
of NN O O
natural NN O I-Task
language NN O I-Task
presuppositions NN O I-Task
that NN O O
is NN O O
presented NN O O
in NN O O
Gazdar NN O O
1979 NN O O
. NN O O

Soames NN O O
1982 NN O O
provides NN O O
a NN O O
theory NN O O
which NN O O
explains NN O O
these NN O O
counterexamples NN O O
. NN O O

Mercer NN O O
1987 NN O O
rejects NN O O
the NN O O
solution NN O O
found NN O O
in NN O O
Soames NN O O
1982 NN O O
leaving NN O O
these NN O O
counterexamples NN O O
unexplained NN O O
. NN O O

By NN O O
reappraising NN O O
these NN O O
insightful NN O O
counterexamples NN O O
, NN O O
the NN O O
inferential NN O O
theory NN O O
for NN O O
natural NN O I-Task
language NN O I-Task
presuppositions NN O I-Task
described NN O O
in NN O O
Mercer NN O O
1987 NN O O
, NN O O
1988 NN O O
gives NN O O
a NN O O
simple NN O O
and NN O O
straightforward NN O O
explanation NN O O
for NN O O
the NN O O
presuppositional NN O I-OtherScientificTerm
nature NN O I-OtherScientificTerm
of NN O O
these NN O O
sentences NN O O
. NN O O

-DOCSTART- NN O O

One NN O O
of NN O O
the NN O O
claimed NN O O
benefits NN O O
of NN O O
Tree NN O I-Method
Adjoining NN O I-Method
Grammars NN O I-Method
is NN O O
that NN O O
they NN O I-Generic
have NN O O
an NN O O
extended NN O I-OtherScientificTerm
domain NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
locality NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
EDOL NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
. NN O O

We NN O O
consider NN O O
how NN O O
this NN O O
can NN O O
be NN O O
exploited NN O O
to NN O O
limit NN O O
the NN O O
need NN O O
for NN O O
feature NN O I-Method
structure NN O I-Method
unification NN O I-Method
during NN O O
parsing NN O I-Task
. NN O O

We NN O O
compare NN O O
two NN O O
wide-coverage NN O O
lexicalized NN O I-Method
grammars NN O I-Method
of NN O I-Method
English NN O I-Method
, NN O O
LEXSYS NN O I-Method
and NN O O
XTAG NN O I-Method
, NN O O
finding NN O O
that NN O O
the NN O O
two NN O O
grammars NN O I-Method
exploit NN O O
EDOL NN O I-OtherScientificTerm
in NN O O
different NN O O
ways NN O O
. NN O O

-DOCSTART- NN O O

Identity NN O I-Task
uncertainty NN O I-Task
is NN O O
a NN O O
pervasive NN O O
problem NN O O
in NN O O
real-world NN O I-Task
data NN O I-Task
analysis NN O I-Task
. NN O O

It NN O O
arises NN O O
whenever NN O O
objects NN O O
are NN O O
not NN O O
labeled NN O O
with NN O O
unique NN O O
identifiers NN O I-Generic
or NN O O
when NN O O
those NN O O
identifiers NN O I-Generic
may NN O O
not NN O O
be NN O O
perceived NN O O
perfectly NN O O
. NN O O

In NN O O
such NN O O
cases NN O O
, NN O O
two NN O O
observations NN O O
may NN O O
or NN O O
may NN O O
not NN O O
correspond NN O O
to NN O O
the NN O O
same NN O O
object NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
consider NN O O
the NN O O
problem NN O I-Generic
in NN O O
the NN O O
context NN O O
of NN O O
citation NN O I-Task
matching NN O I-Task
-- NN O O
the NN O O
problem NN O O
of NN O O
deciding NN O O
which NN O O
citations NN O I-OtherScientificTerm
correspond NN O O
to NN O O
the NN O O
same NN O O
publication NN O I-OtherScientificTerm
. NN O O

Our NN O O
approach NN O I-Generic
is NN O O
based NN O O
on NN O O
the NN O O
use NN O O
of NN O O
a NN O O
relational NN O I-Method
probability NN O I-Method
model NN O I-Method
to NN O O
define NN O O
a NN O O
generative NN O I-Method
model NN O I-Method
for NN O O
the NN O O
domain NN O I-Generic
, NN O O
including NN O O
models NN O I-Method
of NN O I-Method
author NN O I-Method
and NN O I-Method
title NN O I-Method
corruption NN O I-Method
and NN O O
a NN O O
probabilistic NN O I-Method
citation NN O I-Method
grammar NN O I-Method
. NN O O

Identity NN O I-Task
uncertainty NN O I-Task
is NN O O
handled NN O O
by NN O O
extending NN O O
standard NN O O
models NN O I-Generic
to NN O O
incorporate NN O O
probabilities NN O O
over NN O O
the NN O O
possible NN O O
mappings NN O I-OtherScientificTerm
between NN O O
terms NN O O
in NN O O
the NN O O
language NN O O
and NN O O
objects NN O O
in NN O O
the NN O O
domain NN O I-OtherScientificTerm
. NN O O

Inference NN O I-Task
is NN O O
based NN O O
on NN O O
Markov NN O I-Method
chain NN O I-Method
Monte NN O I-Method
Carlo NN O I-Method
, NN O O
augmented NN O O
with NN O O
specific NN O O
methods NN O I-Generic
for NN O O
generating NN O O
efficient NN O O
proposals NN O O
when NN O O
the NN O O
domain NN O O
contains NN O O
many NN O O
objects NN O O
. NN O O

Results NN O O
on NN O O
several NN O O
citation NN O I-Material
data NN O I-Material
sets NN O I-Material
show NN O O
that NN O O
the NN O O
method NN O I-Generic
outperforms NN O O
current NN O I-Method
algorithms NN O I-Method
for NN O O
citation NN O I-Task
matching NN O I-Task
. NN O O

The NN O O
declarative NN O O
, NN O O
relational NN O O
nature NN O O
of NN O O
the NN O O
model NN O I-Generic
also NN O O
means NN O O
that NN O O
our NN O O
algorithm NN O I-Generic
can NN O O
determine NN O O
object NN O I-OtherScientificTerm
characteristics NN O I-OtherScientificTerm
such NN O O
as NN O O
author NN O I-OtherScientificTerm
names NN O I-OtherScientificTerm
by NN O O
combining NN O O
multiple NN O O
citations NN O I-OtherScientificTerm
of NN O O
multiple NN O O
papers NN O O
. NN O O

-DOCSTART- NN O O

The NN O O
paper NN O O
proposes NN O O
and NN O O
empirically NN O O
motivates NN O O
an NN O O
integration NN O O
of NN O O
supervised NN O I-Method
learning NN O I-Method
with NN O O
unsupervised NN O I-Method
learning NN O I-Method
to NN O O
deal NN O O
with NN O O
human NN O I-Task
biases NN O I-Task
in NN O I-Task
summarization NN O I-Task
. NN O O

In NN O O
particular NN O O
, NN O O
we NN O O
explore NN O O
the NN O O
use NN O O
of NN O O
probabilistic NN O I-OtherScientificTerm
decision NN O I-OtherScientificTerm
tree NN O I-OtherScientificTerm
within NN O O
the NN O O
clustering NN O I-Method
framework NN O I-Method
to NN O O
account NN O O
for NN O O
the NN O O
variation NN O O
as NN O O
well NN O O
as NN O O
regularity NN O O
in NN O O
human NN O I-Material
created NN O I-Material
summaries NN O I-Material
. NN O O

The NN O O
corpus NN O I-Material
of NN O I-Material
human NN O I-Material
created NN O I-Material
extracts NN O I-Material
is NN O O
created NN O O
from NN O O
a NN O O
newspaper NN O I-Material
corpus NN O I-Material
and NN O O
used NN O O
as NN O O
a NN O O
test NN O O
set NN O O
. NN O O

We NN O O
build NN O O
probabilistic NN O I-OtherScientificTerm
decision NN O I-OtherScientificTerm
trees NN O I-OtherScientificTerm
of NN O O
different NN O O
flavors NN O O
and NN O O
integrate NN O O
each NN O O
of NN O O
them NN O I-Generic
with NN O O
the NN O O
clustering NN O I-Method
framework NN O I-Method
. NN O O

Experiments NN O O
with NN O O
the NN O O
corpus NN O I-Generic
demonstrate NN O O
that NN O O
the NN O O
mixture NN O O
of NN O O
the NN O O
two NN O O
paradigms NN O O
generally NN O O
gives NN O O
a NN O O
significant NN O O
boost NN O O
in NN O O
performance NN O O
compared NN O O
to NN O O
cases NN O O
where NN O O
either NN O O
ofthe NN O O
two NN O O
is NN O O
considered NN O O
alone NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
study NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
knowledge-independent NN O I-Method
method NN O I-Method
for NN O O
aligning NN O O
terms NN O O
and NN O O
thus NN O O
extracting NN O O
translations NN O O
from NN O O
a NN O O
small NN O I-Material
, NN O I-Material
domain-specific NN O I-Material
corpus NN O I-Material
consisting NN O O
of NN O O
parallel NN O I-Material
English NN O I-Material
and NN O I-Material
Chinese NN O I-Material
court NN O I-Material
judgments NN O I-Material
from NN O O
Hong NN O O
Kong NN O O
. NN O O

With NN O O
a NN O O
sentence-aligned NN O I-Material
corpus NN O I-Material
, NN O O
translation NN O I-OtherScientificTerm
equivalences NN O I-OtherScientificTerm
are NN O O
suggested NN O O
by NN O O
analysing NN O O
the NN O O
frequency NN O I-Generic
profiles NN O I-Generic
of NN O O
parallel NN O I-OtherScientificTerm
concordances NN O I-OtherScientificTerm
. NN O O

The NN O O
method NN O I-Generic
overcomes NN O O
the NN O O
limitations NN O O
of NN O O
conventional NN O O
statistical NN O I-Method
methods NN O I-Method
which NN O O
require NN O O
large NN O I-Material
corpora NN O I-Material
to NN O O
be NN O O
effective NN O O
, NN O O
and NN O O
lexical NN O I-Method
approaches NN O I-Method
which NN O O
depend NN O O
on NN O O
existing NN O O
bilingual NN O I-Material
dictionaries NN O I-Material
. NN O O

Pilot NN O O
testing NN O O
on NN O O
a NN O O
parallel NN O I-Material
corpus NN O I-Material
of NN O O
about NN O O
113K NN O O
Chinese NN O O
words NN O O
and NN O O
120K NN O O
English NN O O
words NN O O
gives NN O O
an NN O O
encouraging NN O O
85 NN O O
% NN O O
precision NN O I-Metric
and NN O O
45 NN O O
% NN O O
recall NN O I-Metric
. NN O O

Future NN O O
work NN O O
includes NN O O
fine-tuning NN O O
the NN O O
algorithm NN O I-Generic
upon NN O O
the NN O O
analysis NN O O
of NN O O
the NN O O
errors NN O O
, NN O O
and NN O O
acquiring NN O O
a NN O O
translation NN O I-OtherScientificTerm
lexicon NN O I-OtherScientificTerm
for NN O O
legal NN O I-Material
terminology NN O I-Material
by NN O O
filtering NN O O
out NN O O
general NN O O
terms NN O O
. NN O O

-DOCSTART- NN O O

Sentiment NN O I-Task
Classification NN O I-Task
seeks NN O O
to NN O O
identify NN O O
a NN O O
piece NN O O
of NN O O
text NN O O
according NN O O
to NN O O
its NN O O
author NN O O
's NN O O
general NN O O
feeling NN O O
toward NN O O
their NN O O
subject NN O O
, NN O O
be NN O O
it NN O O
positive NN O O
or NN O O
negative NN O O
. NN O O

Traditional NN O O
machine NN O I-Method
learning NN O I-Method
techniques NN O I-Method
have NN O O
been NN O O
applied NN O O
to NN O O
this NN O O
problem NN O I-Generic
with NN O O
reasonable NN O O
success NN O O
, NN O O
but NN O O
they NN O O
have NN O O
been NN O O
shown NN O O
to NN O O
work NN O O
well NN O O
only NN O O
when NN O O
there NN O O
is NN O O
a NN O O
good NN O O
match NN O O
between NN O O
the NN O O
training NN O I-Material
and NN O I-Material
test NN O I-Material
data NN O I-Material
with NN O O
respect NN O O
to NN O O
topic NN O O
. NN O O

This NN O O
paper NN O O
demonstrates NN O O
that NN O O
match NN O O
with NN O O
respect NN O O
to NN O O
domain NN O O
and NN O O
time NN O O
is NN O O
also NN O O
important NN O O
, NN O O
and NN O O
presents NN O O
preliminary NN O O
experiments NN O O
with NN O O
training NN O I-Material
data NN O I-Material
labeled NN O O
with NN O O
emoticons NN O I-OtherScientificTerm
, NN O O
which NN O O
has NN O O
the NN O O
potential NN O O
of NN O O
being NN O O
independent NN O O
of NN O O
domain NN O O
, NN O O
topic NN O O
and NN O O
time NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
a NN O O
novel NN O O
algorithm NN O I-Generic
for NN O O
estimating NN O O
the NN O O
broad NN O O
3D NN O I-Task
geometric NN O I-Task
structure NN O I-Task
of NN O I-Task
outdoor NN O I-Task
video NN O I-Task
scenes NN O I-Task
. NN O O

Leveraging NN O O
spatio-temporal NN O I-Method
video NN O I-Method
segmentation NN O I-Method
, NN O O
we NN O O
decompose NN O O
a NN O O
dynamic NN O I-OtherScientificTerm
scene NN O I-OtherScientificTerm
captured NN O O
by NN O O
a NN O O
video NN O O
into NN O O
geometric NN O I-OtherScientificTerm
classes NN O I-OtherScientificTerm
, NN O O
based NN O O
on NN O O
predictions NN O O
made NN O O
by NN O O
region-classifiers NN O I-Method
that NN O O
are NN O O
trained NN O O
on NN O O
appearance NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
motion NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
. NN O O

By NN O O
examining NN O O
the NN O O
homogeneity NN O O
of NN O O
the NN O O
prediction NN O O
, NN O O
we NN O O
combine NN O O
predictions NN O O
across NN O O
multiple NN O O
segmentation NN O I-OtherScientificTerm
hierarchy NN O I-OtherScientificTerm
levels NN O I-OtherScientificTerm
alleviating NN O O
the NN O O
need NN O O
to NN O O
determine NN O O
the NN O O
granularity NN O I-OtherScientificTerm
a NN O I-OtherScientificTerm
priori NN O I-OtherScientificTerm
. NN O O

We NN O O
built NN O O
a NN O O
novel NN O O
, NN O O
extensive NN O O
dataset NN O I-Generic
on NN O O
geometric NN O I-OtherScientificTerm
context NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
video NN O I-OtherScientificTerm
to NN O O
evaluate NN O O
our NN O O
method NN O I-Generic
, NN O O
consisting NN O O
of NN O O
over NN O O
100 NN O O
ground-truth NN O O
annotated NN O I-OtherScientificTerm
outdoor NN O I-OtherScientificTerm
videos NN O I-OtherScientificTerm
with NN O O
over NN O O
20,000 NN O O
frames NN O O
. NN O O

To NN O O
further NN O O
scale NN O O
beyond NN O O
this NN O O
dataset NN O I-Generic
, NN O O
we NN O O
propose NN O O
a NN O O
semi-supervised NN O I-Method
learning NN O I-Method
framework NN O I-Method
to NN O O
expand NN O O
the NN O O
pool NN O O
of NN O O
labeled NN O I-Material
data NN O I-Material
with NN O O
high NN O I-OtherScientificTerm
confidence NN O I-OtherScientificTerm
predictions NN O I-OtherScientificTerm
obtained NN O O
from NN O O
unlabeled NN O I-OtherScientificTerm
data NN O I-OtherScientificTerm
. NN O O

Our NN O O
system NN O I-Generic
produces NN O O
an NN O O
accurate NN O O
prediction NN O O
of NN O O
geometric NN O I-OtherScientificTerm
context NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
video NN O I-OtherScientificTerm
achieving NN O O
96 NN O O
% NN O O
accuracy NN O I-Metric
across NN O O
main NN O O
geometric NN O O
classes NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
a NN O O
system NN O I-Generic
-LRB- NN O O
RAREAS NN O I-Method
-RRB- NN O O
which NN O O
synthesizes NN O O
marine NN O I-Task
weather NN O I-Task
forecasts NN O I-Task
directly NN O O
from NN O O
formatted NN O I-Material
weather NN O I-Material
data NN O I-Material
. NN O O

Such NN O O
synthesis NN O I-Generic
appears NN O O
feasible NN O O
in NN O O
certain NN O O
natural NN O I-Material
sublanguages NN O I-Material
with NN O I-Material
stereotyped NN O I-Material
text NN O I-Material
structure NN O I-Material
. NN O O

RAREAS NN O I-Method
draws NN O O
on NN O O
several NN O O
kinds NN O O
of NN O O
linguistic NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
non-linguistic NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
and NN O O
mirrors NN O O
a NN O O
forecaster NN O O
's NN O O
apparent NN O O
tendency NN O O
to NN O O
ascribe NN O O
less NN O O
precise NN O O
temporal NN O O
adverbs NN O O
to NN O O
more NN O O
remote NN O O
meteorological NN O O
events NN O O
. NN O O

The NN O O
approach NN O I-Generic
can NN O O
easily NN O O
be NN O O
adapted NN O O
to NN O O
synthesize NN O O
bilingual NN O I-Material
or NN O I-Material
multi-lingual NN O I-Material
texts NN O I-Material
. NN O O

-DOCSTART- NN O O

When NN O O
people NN O O
use NN O O
natural NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
in NN O O
natural NN O O
settings NN O O
, NN O O
they NN O O
often NN O O
use NN O O
it NN O I-Generic
ungrammatically NN O O
, NN O O
missing NN O O
out NN O O
or NN O O
repeating NN O O
words NN O O
, NN O O
breaking-off NN O O
and NN O O
restarting NN O O
, NN O O
speaking NN O O
in NN O O
fragments NN O O
, NN O O
etc. NN O O
. NN O O

Their NN O O
human NN O O
listeners NN O O
are NN O O
usually NN O O
able NN O O
to NN O O
cope NN O O
with NN O O
these NN O O
deviations NN O O
with NN O O
little NN O O
difficulty NN O O
. NN O O

If NN O O
a NN O O
computer NN O I-Method
system NN O I-Method
wishes NN O O
to NN O O
accept NN O O
natural NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
input NN O I-OtherScientificTerm
from NN O O
its NN O O
users NN O O
on NN O O
a NN O O
routine NN O O
basis NN O O
, NN O O
it NN O O
must NN O O
display NN O O
a NN O O
similar NN O O
indifference NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
outline NN O O
a NN O O
set NN O O
of NN O O
parsing NN O I-OtherScientificTerm
flexibilities NN O I-OtherScientificTerm
that NN O O
such NN O O
a NN O O
system NN O I-Generic
should NN O O
provide NN O O
. NN O O

We NN O O
go NN O O
, NN O O
on NN O O
to NN O O
describe NN O O
FlexP NN O I-Method
, NN O O
a NN O O
bottom-up NN O I-Method
pattern-matching NN O I-Method
parser NN O I-Method
that NN O O
we NN O O
have NN O O
designed NN O O
and NN O O
implemented NN O O
to NN O O
provide NN O O
these NN O O
flexibilities NN O I-Generic
for NN O O
restricted NN O I-OtherScientificTerm
natural NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
input NN O O
to NN O O
a NN O O
limited-domain NN O I-Method
computer NN O I-Method
system NN O I-Method
. NN O O

-DOCSTART- NN O O

Oral NN O I-Material
communication NN O I-Material
is NN O O
ubiquitous NN O O
and NN O O
carries NN O O
important NN O O
information NN O O
yet NN O O
it NN O O
is NN O O
also NN O O
time NN O O
consuming NN O O
to NN O O
document NN O O
. NN O O

Given NN O O
the NN O O
development NN O O
of NN O O
storage NN O I-Task
media NN O I-Task
and NN O I-Task
networks NN O I-Task
one NN O O
could NN O O
just NN O O
record NN O O
and NN O O
store NN O O
a NN O O
conversation NN O I-Material
for NN O O
documentation NN O O
. NN O O

The NN O O
question NN O O
is NN O O
, NN O O
however NN O O
, NN O O
how NN O O
an NN O O
interesting NN O O
information NN O O
piece NN O O
would NN O O
be NN O O
found NN O O
in NN O O
a NN O O
large NN O O
database NN O O
. NN O O

Traditional NN O O
information NN O I-Method
retrieval NN O I-Method
techniques NN O I-Method
use NN O O
a NN O O
histogram NN O I-Method
of NN O I-Method
keywords NN O I-Method
as NN O O
the NN O O
document NN O I-Method
representation NN O I-Method
but NN O O
oral NN O I-Material
communication NN O I-Material
may NN O O
offer NN O O
additional NN O O
indices NN O O
such NN O O
as NN O O
the NN O O
time NN O O
and NN O O
place NN O O
of NN O O
the NN O O
rejoinder NN O O
and NN O O
the NN O O
attendance NN O O
. NN O O

An NN O O
alternative NN O O
index NN O O
could NN O O
be NN O O
the NN O O
activity NN O I-OtherScientificTerm
such NN O O
as NN O O
discussing NN O I-OtherScientificTerm
, NN O O
planning NN O I-OtherScientificTerm
, NN O O
informing NN O I-OtherScientificTerm
, NN O O
story-telling NN O I-OtherScientificTerm
, NN O O
etc. NN O O
. NN O O

This NN O O
paper NN O O
addresses NN O O
the NN O O
problem NN O O
of NN O O
the NN O O
automatic NN O I-Task
detection NN O I-Task
of NN O O
those NN O O
activities NN O I-Generic
in NN O O
meeting NN O O
situation NN O O
and NN O O
everyday NN O O
rejoinders NN O O
. NN O O

Several NN O O
extensions NN O O
of NN O O
this NN O O
basic NN O O
idea NN O O
are NN O O
being NN O O
discussed NN O O
and/or NN O O
evaluated NN O O
: NN O O
Similar NN O O
to NN O O
activities NN O I-Generic
one NN O O
can NN O O
define NN O O
subsets NN O O
of NN O O
larger NN O O
database NN O O
and NN O O
detect NN O O
those NN O O
automatically NN O O
which NN O O
is NN O O
shown NN O O
on NN O O
a NN O O
large NN O O
database NN O I-Material
of NN O I-Material
TV NN O I-Material
shows NN O I-Material
. NN O O

Emotions NN O I-Material
and NN O O
other NN O O
indices NN O O
such NN O O
as NN O O
the NN O O
dominance NN O I-OtherScientificTerm
distribution NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
speakers NN O I-OtherScientificTerm
might NN O O
be NN O O
available NN O O
on NN O O
the NN O O
surface NN O O
and NN O O
could NN O O
be NN O O
used NN O O
directly NN O O
. NN O O

Despite NN O O
the NN O O
small NN O O
size NN O O
of NN O O
the NN O O
databases NN O O
used NN O O
some NN O O
results NN O O
about NN O O
the NN O O
effectiveness NN O O
of NN O O
these NN O O
indices NN O O
can NN O O
be NN O O
obtained NN O O
. NN O O

-DOCSTART- NN O O

Taiwan NN O I-Material
Child NN O I-Material
Language NN O I-Material
Corpus NN O I-Material
contains NN O O
scripts NN O O
transcribed NN O O
from NN O O
about NN O O
330 NN O O
hours NN O O
of NN O O
recordings NN O O
of NN O O
fourteen NN O O
young NN O O
children NN O O
from NN O O
Southern NN O O
Min NN O O
Chinese NN O O
speaking NN O O
families NN O O
in NN O O
Taiwan NN O O
. NN O O

The NN O O
format NN O O
of NN O O
the NN O O
corpus NN O I-Generic
adopts NN O O
the NN O O
Child NN O I-Material
Language NN O I-Material
Data NN O I-Material
Exchange NN O I-Material
System NN O I-Material
-LRB- NN O I-Material
CHILDES NN O I-Material
-RRB- NN O I-Material
. NN O O

The NN O O
size NN O O
of NN O O
the NN O O
corpus NN O I-Generic
is NN O O
about NN O O
1.6 NN O O
million NN O O
words NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
describe NN O O
data NN O I-Task
collection NN O I-Task
, NN O O
transcription NN O I-Task
, NN O O
word NN O I-Task
segmentation NN O I-Task
, NN O O
and NN O O
part-of-speech NN O I-Task
annotation NN O I-Task
of NN O O
this NN O O
corpus NN O I-Generic
. NN O O

Applications NN O O
of NN O O
the NN O O
corpus NN O I-Generic
are NN O O
also NN O O
discussed NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
shows NN O O
how NN O O
dictionary NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
sense NN O I-OtherScientificTerm
definitions NN O I-OtherScientificTerm
can NN O O
be NN O O
analysed NN O O
by NN O O
applying NN O O
a NN O O
hierarchy NN O I-Method
of NN O I-Method
phrasal NN O I-Method
patterns NN O I-Method
. NN O O

An NN O O
experimental NN O O
system NN O I-Generic
embodying NN O O
this NN O O
mechanism NN O I-Generic
has NN O O
been NN O O
implemented NN O O
for NN O O
processing NN O O
definitions NN O O
from NN O O
the NN O O
Longman NN O I-Material
Dictionary NN O I-Material
of NN O I-Material
Contemporary NN O I-Material
English NN O I-Material
. NN O O

A NN O O
property NN O O
of NN O O
this NN O O
dictionary NN O I-Generic
, NN O O
exploited NN O O
by NN O O
the NN O O
system NN O I-Generic
, NN O O
is NN O O
that NN O O
it NN O I-Generic
uses NN O O
a NN O O
restricted NN O I-OtherScientificTerm
vocabulary NN O I-OtherScientificTerm
in NN O O
its NN O O
word NN O I-OtherScientificTerm
sense NN O I-OtherScientificTerm
definitions NN O I-OtherScientificTerm
. NN O O

The NN O O
structures NN O O
generated NN O O
by NN O O
the NN O O
experimental NN O O
system NN O O
are NN O O
intended NN O O
to NN O O
be NN O O
used NN O O
for NN O O
the NN O O
classification NN O I-Task
of NN O I-Task
new NN O I-Task
word NN O I-Task
senses NN O I-Task
in NN O O
terms NN O O
of NN O O
the NN O O
senses NN O O
of NN O O
words NN O O
in NN O O
the NN O O
restricted NN O I-OtherScientificTerm
vocabulary NN O I-OtherScientificTerm
. NN O O

Examples NN O O
illustrating NN O O
the NN O O
output NN O O
generated NN O O
are NN O O
presented NN O O
, NN O O
and NN O O
some NN O O
qualitative NN O O
performance NN O O
results NN O O
and NN O O
problems NN O O
that NN O O
were NN O O
encountered NN O O
are NN O O
discussed NN O O
. NN O O

The NN O O
analysis NN O O
process NN O O
applies NN O O
successively NN O O
more NN O O
specific NN O O
phrasal NN O I-OtherScientificTerm
analysis NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
as NN O O
determined NN O O
by NN O O
a NN O O
hierarchy NN O O
of NN O O
patterns NN O O
in NN O O
which NN O O
less NN O O
specific NN O O
patterns NN O O
dominate NN O O
more NN O O
specific NN O O
ones NN O O
. NN O O

This NN O O
ensures NN O O
that NN O O
reasonable NN O O
incomplete NN O O
analyses NN O O
of NN O O
the NN O O
definitions NN O O
are NN O O
produced NN O O
when NN O O
more NN O O
complete NN O O
analyses NN O O
are NN O O
not NN O O
possible NN O O
, NN O O
resulting NN O O
in NN O O
a NN O O
relatively NN O O
robust NN O O
analysis NN O O
mechanism NN O O
. NN O O

Thus NN O O
the NN O O
work NN O O
reported NN O O
addresses NN O O
two NN O O
robustness NN O I-Task
problems NN O I-Task
faced NN O O
by NN O O
current NN O O
experimental NN O O
natural NN O I-Method
language NN O I-Method
processing NN O I-Method
systems NN O I-Method
: NN O O
coping NN O O
with NN O O
an NN O O
incomplete NN O I-Material
lexicon NN O I-Material
and NN O O
with NN O O
incomplete NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
phrasal NN O I-OtherScientificTerm
constructions NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
word NN O I-Method
segmentation NN O I-Method
system NN O I-Method
in NN O O
France NN O O
Telecom NN O O
R&D NN O O
Beijing NN O O
, NN O O
which NN O O
uses NN O O
a NN O O
unified NN O O
approach NN O I-Generic
to NN O O
word NN O I-Task
breaking NN O I-Task
and NN O O
OOV NN O I-Task
identification NN O I-Task
. NN O O

The NN O O
output NN O O
can NN O O
be NN O O
customized NN O O
to NN O O
meet NN O O
different NN O O
segmentation NN O O
standards NN O O
through NN O O
the NN O O
application NN O O
of NN O O
an NN O O
ordered NN O O
list NN O O
of NN O O
transformation NN O O
. NN O O

The NN O O
system NN O I-Generic
participated NN O O
in NN O O
all NN O O
the NN O O
tracks NN O O
of NN O O
the NN O O
segmentation NN O I-OtherScientificTerm
bakeoff NN O I-OtherScientificTerm
-- NN O O
PK-open NN O I-OtherScientificTerm
, NN O O
PK-closed NN O I-OtherScientificTerm
, NN O O
AS-open NN O I-OtherScientificTerm
, NN O O
AS-closed NN O I-OtherScientificTerm
, NN O O
HK-open NN O I-OtherScientificTerm
, NN O O
HK-closed NN O I-OtherScientificTerm
, NN O O
MSR-open NN O I-OtherScientificTerm
and NN O O
MSR NN O I-OtherScientificTerm
- NN O I-OtherScientificTerm
closed NN O I-OtherScientificTerm
-- NN O O
and NN O O
achieved NN O O
the NN O O
state-of-the-art NN O O
performance NN O O
in NN O O
MSR-open NN O I-OtherScientificTerm
, NN O O
MSR-close NN O I-OtherScientificTerm
and NN O O
PK-open NN O I-OtherScientificTerm
tracks NN O O
. NN O O

Analysis NN O O
of NN O O
the NN O O
results NN O O
shows NN O O
that NN O O
each NN O O
component NN O O
of NN O O
the NN O O
system NN O I-Generic
contributed NN O O
to NN O O
the NN O O
scores NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
a NN O O
method NN O I-Generic
of NN O O
interactively NN O I-Task
visualizing NN O I-Task
and NN O I-Task
directing NN O I-Task
the NN O I-Task
process NN O I-Task
of NN O I-Task
translating NN O I-Task
a NN O O
sentence NN O O
. NN O O

The NN O O
method NN O I-Generic
allows NN O O
a NN O O
user NN O O
to NN O O
explore NN O O
a NN O O
model NN O I-Generic
of NN O O
syntax-based NN O I-Task
statistical NN O I-Task
machine NN O I-Task
translation NN O I-Task
-LRB- NN O I-Task
MT NN O I-Task
-RRB- NN O I-Task
, NN O O
to NN O O
understand NN O O
the NN O O
model NN O I-Generic
's NN O O
strengths NN O O
and NN O O
weaknesses NN O O
, NN O O
and NN O O
to NN O O
compare NN O O
it NN O I-Generic
to NN O O
other NN O O
MT NN O I-Method
systems NN O I-Method
. NN O O

Using NN O O
this NN O O
visualization NN O I-Method
method NN O I-Method
, NN O O
we NN O O
can NN O O
find NN O O
and NN O O
address NN O O
conceptual NN O O
and NN O O
practical NN O O
problems NN O O
in NN O O
an NN O O
MT NN O I-Method
system NN O I-Method
. NN O O

In NN O O
our NN O O
demonstration NN O O
at NN O O
ACL NN O O
, NN O O
new NN O O
users NN O O
of NN O O
our NN O O
tool NN O O
will NN O O
drive NN O O
a NN O O
syntax-based NN O I-Method
decoder NN O I-Method
for NN O O
themselves NN O O
. NN O O

-DOCSTART- NN O O

A NN O O
method NN O I-Generic
of NN O O
sense NN O I-Task
resolution NN O I-Task
is NN O O
proposed NN O O
that NN O O
is NN O O
based NN O O
on NN O O
WordNet NN O I-Material
, NN O O
an NN O O
on-line NN O I-Material
lexical NN O I-Material
database NN O I-Material
that NN O O
incorporates NN O O
semantic NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
-LRB- NN O O
synonymy NN O I-OtherScientificTerm
, NN O O
antonymy NN O I-OtherScientificTerm
, NN O O
hyponymy NN O I-OtherScientificTerm
, NN O O
meronymy NN O I-OtherScientificTerm
, NN O O
causal NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
troponymic NN O I-OtherScientificTerm
entailment NN O I-OtherScientificTerm
-RRB- NN O O
as NN O O
labeled NN O O
pointers NN O O
between NN O O
word NN O O
senses NN O O
. NN O O

With NN O O
WordNet NN O I-Material
, NN O O
it NN O O
is NN O O
easy NN O O
to NN O O
retrieve NN O O
sets NN O O
of NN O O
semantically NN O I-OtherScientificTerm
related NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
, NN O O
a NN O O
facility NN O O
that NN O O
will NN O O
be NN O O
used NN O O
for NN O O
sense NN O I-Task
resolution NN O I-Task
during NN O O
text NN O I-Task
processing NN O I-Task
, NN O O
as NN O O
follows NN O O
. NN O O

When NN O O
a NN O O
word NN O O
with NN O O
multiple NN O O
senses NN O O
is NN O O
encountered NN O O
, NN O O
one NN O O
of NN O O
two NN O O
procedures NN O I-Generic
will NN O O
be NN O O
followed NN O O
. NN O O

Either NN O O
, NN O O
-LRB- NN O O
1 NN O O
-RRB- NN O O
words NN O O
related NN O O
in NN O O
meaning NN O O
to NN O O
the NN O O
alternative NN O I-OtherScientificTerm
senses NN O I-OtherScientificTerm
of NN O O
the NN O O
polysemous NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
will NN O O
be NN O O
retrieved NN O O
; NN O O
new NN O O
strings NN O O
will NN O O
be NN O O
derived NN O O
by NN O O
substituting NN O O
these NN O O
related NN O O
words NN O O
into NN O O
the NN O O
context NN O O
of NN O O
the NN O O
polysemous NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
; NN O O
a NN O O
large NN O O
textual NN O O
corpus NN O O
will NN O O
then NN O O
be NN O O
searched NN O O
for NN O O
these NN O O
derived NN O O
strings NN O O
; NN O O
and NN O O
that NN O O
sense NN O O
will NN O O
be NN O O
chosen NN O O
that NN O O
corresponds NN O O
to NN O O
the NN O O
derived NN O O
string NN O O
that NN O O
is NN O O
found NN O O
most NN O O
often NN O O
in NN O O
the NN O O
corpus NN O O
. NN O O

Or NN O O
, NN O O
-LRB- NN O O
2 NN O O
-RRB- NN O O
the NN O O
context NN O O
of NN O O
the NN O O
polysemous NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
will NN O O
be NN O O
used NN O O
as NN O O
a NN O O
key NN O O
to NN O O
search NN O O
a NN O O
large NN O O
corpus NN O O
; NN O O
all NN O O
words NN O O
found NN O O
to NN O O
occur NN O O
in NN O O
that NN O O
context NN O O
will NN O O
be NN O O
noted NN O O
; NN O O
WordNet NN O I-Material
will NN O O
then NN O O
be NN O O
used NN O O
to NN O O
estimate NN O O
the NN O O
semantic NN O I-OtherScientificTerm
distance NN O I-OtherScientificTerm
from NN O O
those NN O O
words NN O O
to NN O O
the NN O O
alternative NN O O
senses NN O O
of NN O O
the NN O O
polysemous NN O O
word NN O O
; NN O O
and NN O O
that NN O O
sense NN O O
will NN O O
be NN O O
chosen NN O O
that NN O O
is NN O O
closest NN O O
in NN O O
meaning NN O O
to NN O O
other NN O O
words NN O O
occurring NN O O
in NN O O
the NN O O
same NN O O
context NN O O
If NN O O
successful NN O O
, NN O O
this NN O O
procedure NN O I-Generic
could NN O O
have NN O O
practical NN O O
applications NN O O
to NN O O
problems NN O O
of NN O O
information NN O I-Task
retrieval NN O I-Task
, NN O O
mechanical NN O I-Task
translation NN O I-Task
, NN O O
intelligent NN O I-Task
tutoring NN O I-Task
systems NN O I-Task
, NN O O
and NN O O
elsewhere NN O O
. NN O O

-DOCSTART- NN O O

The NN O O
interlingual NN O I-Method
approach NN O I-Method
to NN O O
MT NN O I-Generic
has NN O O
been NN O O
repeatedly NN O O
advocated NN O O
by NN O O
researchers NN O O
originally NN O O
interested NN O O
in NN O O
natural NN O I-Task
language NN O I-Task
understanding NN O I-Task
who NN O O
take NN O O
machine NN O I-Task
translation NN O I-Task
to NN O O
be NN O O
one NN O O
possible NN O O
application NN O O
. NN O O

However NN O O
, NN O O
not NN O O
only NN O O
the NN O O
ambiguity NN O O
but NN O O
also NN O O
the NN O O
vagueness NN O O
which NN O O
every NN O O
natural NN O O
language NN O O
inevitably NN O O
has NN O O
leads NN O O
this NN O O
approach NN O O
into NN O O
essential NN O O
difficulties NN O O
. NN O O

In NN O O
contrast NN O O
, NN O O
our NN O O
project NN O O
, NN O O
the NN O O
Mu-project NN O I-Method
, NN O O
adopts NN O O
the NN O O
transfer NN O I-Method
approach NN O I-Method
as NN O O
the NN O O
basic NN O O
framework NN O O
of NN O O
MT NN O I-Task
. NN O O

This NN O O
paper NN O O
describes NN O O
the NN O O
detailed NN O O
construction NN O O
of NN O O
the NN O O
transfer NN O I-Generic
phase NN O I-Generic
of NN O O
our NN O O
system NN O I-Generic
from NN O O
Japanese NN O I-Material
to NN O O
English NN O I-Material
, NN O O
and NN O O
gives NN O O
some NN O O
examples NN O O
of NN O O
problems NN O O
which NN O O
seem NN O O
difficult NN O O
to NN O O
treat NN O O
in NN O O
the NN O O
interlingual NN O I-Method
approach NN O I-Method
. NN O O

The NN O O
basic NN O O
design NN O O
principles NN O O
of NN O O
the NN O O
transfer NN O I-Generic
phase NN O I-Generic
of NN O O
our NN O O
system NN O I-Generic
have NN O O
already NN O O
been NN O O
mentioned NN O O
in NN O O
-LRB- NN O O
1 NN O O
-RRB- NN O O
-LRB- NN O O
2 NN O O
-RRB- NN O O
. NN O O

Some NN O O
of NN O O
the NN O O
principles NN O I-Generic
which NN O O
are NN O O
relevant NN O O
to NN O O
the NN O O
topic NN O O
of NN O O
this NN O O
paper NN O O
are NN O O
: NN O O
-LRB- NN O O
a NN O O
-RRB- NN O O
Multiple NN O I-Method
Layer NN O I-Method
of NN O I-Method
Grammars NN O I-Method
-LRB- NN O O
b NN O O
-RRB- NN O O
Multiple NN O I-Method
Layer NN O I-Method
Presentation NN O I-Method
-LRB- NN O O
c NN O O
-RRB- NN O O
Lexicon NN O I-Method
Driven NN O I-Method
Processing NN O I-Method
-LRB- NN O O
d NN O O
-RRB- NN O O
Form-Oriented NN O I-Method
Dictionary NN O I-Method
Description NN O I-Method
. NN O O

This NN O O
paper NN O O
also NN O O
shows NN O O
how NN O O
these NN O O
principles NN O I-Generic
are NN O O
realized NN O O
in NN O O
the NN O O
current NN O O
system NN O I-Generic
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
discourse NN O I-OtherScientificTerm
segments NN O I-OtherScientificTerm
are NN O O
defined NN O O
and NN O O
a NN O O
method NN O I-Generic
for NN O O
discourse NN O I-Task
segmentation NN O I-Task
primarily NN O O
based NN O O
on NN O O
abduction NN O I-Task
of NN O I-Task
temporal NN O I-Task
relations NN O I-Task
between NN O O
segments NN O O
is NN O O
proposed NN O O
. NN O O

This NN O O
method NN O I-Generic
is NN O O
precise NN O O
and NN O O
computationally NN O O
feasible NN O O
and NN O O
is NN O O
supported NN O O
by NN O O
previous NN O O
work NN O O
in NN O O
the NN O O
area NN O O
of NN O O
temporal NN O I-Task
anaphora NN O I-Task
resolution NN O I-Task
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
to NN O O
what NN O O
extent NN O O
deep NN O I-Task
processing NN O I-Task
may NN O O
benefit NN O O
from NN O O
shallow NN O I-Method
techniques NN O I-Method
and NN O O
it NN O O
presents NN O O
a NN O O
NLP NN O I-Method
system NN O I-Method
which NN O O
integrates NN O O
a NN O O
linguistic NN O I-Method
PoS NN O I-Method
tagger NN O I-Method
and NN O I-Method
chunker NN O I-Method
as NN O O
a NN O O
preprocessing NN O O
module NN O O
of NN O O
a NN O O
broad NN O I-Method
coverage NN O I-Method
unification NN O I-Method
based NN O I-Method
grammar NN O I-Method
of NN O I-Method
Spanish NN O I-Method
. NN O O

Experiments NN O O
show NN O O
that NN O O
the NN O O
efficiency NN O O
of NN O O
the NN O O
overall NN O O
analysis NN O O
improves NN O O
significantly NN O O
and NN O O
that NN O O
our NN O O
system NN O I-Generic
also NN O O
provides NN O O
robustness NN O I-Metric
to NN O O
the NN O O
linguistic NN O I-Task
processing NN O I-Task
while NN O O
maintaining NN O O
both NN O O
the NN O O
accuracy NN O I-Metric
and NN O O
the NN O O
precision NN O I-Metric
of NN O O
the NN O O
grammar NN O O
. NN O O

-DOCSTART- NN O O

Joint NN O I-Method
image NN O I-Method
filters NN O I-Method
can NN O O
leverage NN O O
the NN O O
guidance NN O I-Material
image NN O I-Material
as NN O O
a NN O O
prior NN O O
and NN O O
transfer NN O O
the NN O O
structural NN O O
details NN O O
from NN O O
the NN O O
guidance NN O I-Material
image NN O I-Material
to NN O O
the NN O O
target NN O O
image NN O O
for NN O O
suppressing NN O I-Task
noise NN O I-Task
or NN O O
enhancing NN O I-Task
spatial NN O I-Task
resolution NN O I-Task
. NN O O

Existing NN O I-Generic
methods NN O I-Generic
rely NN O O
on NN O O
various NN O O
kinds NN O O
of NN O O
explicit NN O I-Method
filter NN O I-Method
construction NN O I-Method
or NN O O
hand-designed NN O I-Method
objective NN O I-Method
functions NN O I-Method
. NN O O

It NN O O
is NN O O
thus NN O O
difficult NN O O
to NN O O
understand NN O O
, NN O O
improve NN O O
, NN O O
and NN O O
accelerate NN O O
them NN O I-Generic
in NN O O
a NN O O
coherent NN O I-Generic
framework NN O I-Generic
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
learning-based NN O I-Method
approach NN O I-Method
to NN O O
construct NN O O
a NN O O
joint NN O I-Method
filter NN O I-Method
based NN O O
on NN O O
Convolution-al NN O I-Method
Neural NN O I-Method
Networks NN O I-Method
. NN O O

In NN O O
contrast NN O O
to NN O O
existing NN O O
methods NN O I-Generic
that NN O O
consider NN O O
only NN O O
the NN O O
guidance NN O I-Material
image NN O I-Material
, NN O O
our NN O O
method NN O I-Generic
can NN O O
selectively NN O O
transfer NN O I-Task
salient NN O I-Task
structures NN O I-Task
that NN O O
are NN O O
consistent NN O O
in NN O O
both NN O O
guidance NN O O
and NN O O
target NN O O
images NN O O
. NN O O

We NN O O
show NN O O
that NN O O
the NN O O
model NN O I-Generic
trained NN O O
on NN O O
a NN O O
certain NN O O
type NN O O
of NN O O
data NN O I-Generic
, NN O O
e.g. NN O O
, NN O O
RGB NN O I-Material
and NN O I-Material
depth NN O I-Material
images NN O I-Material
, NN O O
generalizes NN O O
well NN O O
for NN O O
other NN O O
modalities NN O I-Generic
, NN O O
e.g. NN O O
, NN O O
Flash/Non-Flash NN O I-Material
and NN O I-Material
RGB/NIR NN O I-Material
images NN O I-Material
. NN O O

We NN O O
validate NN O O
the NN O O
effectiveness NN O O
of NN O O
the NN O O
proposed NN O O
joint NN O I-Method
filter NN O I-Method
through NN O O
extensive NN O O
comparisons NN O O
with NN O O
state-of-the-art NN O I-Generic
methods NN O I-Generic
. NN O O

-DOCSTART- NN O O

In NN O O
our NN O O
current NN O O
research NN O O
into NN O O
the NN O O
design NN O O
of NN O O
cognitively NN O I-Task
well-motivated NN O I-Task
interfaces NN O I-Task
relying NN O O
primarily NN O O
on NN O O
the NN O O
display NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
graphical NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
, NN O O
we NN O O
have NN O O
observed NN O O
that NN O O
graphical NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
alone NN O O
does NN O O
not NN O O
provide NN O O
sufficient NN O O
support NN O O
to NN O O
users NN O O
- NN O O
particularly NN O O
when NN O O
situations NN O O
arise NN O O
that NN O O
do NN O O
not NN O O
simply NN O O
conform NN O O
to NN O O
the NN O O
users NN O O
' NN O O
expectations NN O O
. NN O O

This NN O O
can NN O O
occur NN O O
due NN O O
to NN O O
too NN O O
much NN O O
information NN O O
being NN O O
requested NN O O
, NN O O
too NN O O
little NN O O
, NN O O
information NN O O
of NN O O
the NN O O
wrong NN O O
kind NN O O
, NN O O
etc. NN O O
. NN O O

To NN O O
solve NN O O
this NN O O
problem NN O O
, NN O O
we NN O O
are NN O O
working NN O O
towards NN O O
the NN O O
integration NN O O
of NN O O
natural NN O I-Task
language NN O I-Task
generation NN O I-Task
to NN O O
augment NN O O
the NN O O
interaction NN O I-Generic

-DOCSTART- NN O O

A NN O O
central NN O O
problem NN O O
of NN O O
word NN O I-Task
sense NN O I-Task
disambiguation NN O I-Task
-LRB- NN O I-Task
WSD NN O I-Task
-RRB- NN O I-Task
is NN O O
the NN O O
lack NN O O
of NN O O
manually NN O I-Material
sense-tagged NN O I-Material
data NN O I-Material
required NN O O
for NN O O
supervised NN O I-Method
learning NN O I-Method
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
evaluate NN O O
an NN O O
approach NN O I-Generic
to NN O O
automatically NN O O
acquire NN O O
sense-tagged NN O I-Material
training NN O I-Material
data NN O I-Material
from NN O O
English-Chinese NN O I-Material
parallel NN O I-Material
corpora NN O I-Material
, NN O O
which NN O O
are NN O O
then NN O O
used NN O O
for NN O O
disambiguating NN O O
the NN O O
nouns NN O I-OtherScientificTerm
in NN O O
the NN O O
SENSEVAL-2 NN O I-Material
English NN O I-Material
lexical NN O I-Material
sample NN O I-Material
task NN O I-Material
. NN O O

Our NN O O
investigation NN O O
reveals NN O O
that NN O O
this NN O O
method NN O I-Generic
of NN O O
acquiring NN O I-Task
sense-tagged NN O I-Task
data NN O I-Task
is NN O O
promising NN O O
. NN O O

On NN O O
a NN O O
subset NN O O
of NN O O
the NN O O
most NN O O
difficult NN O O
SENSEVAL-2 NN O I-OtherScientificTerm
nouns NN O I-OtherScientificTerm
, NN O O
the NN O O
accuracy NN O I-Metric
difference NN O O
between NN O O
the NN O O
two NN O O
approaches NN O O
is NN O O
only NN O O
14.0 NN O O
% NN O O
, NN O O
and NN O O
the NN O O
difference NN O O
could NN O O
narrow NN O O
further NN O O
to NN O O
6.5 NN O O
% NN O O
if NN O O
we NN O O
disregard NN O O
the NN O O
advantage NN O O
that NN O O
manually NN O I-Material
sense-tagged NN O I-Material
data NN O I-Material
have NN O O
in NN O O
their NN O O
sense NN O I-Metric
coverage NN O I-Metric
. NN O O

Our NN O O
analysis NN O O
also NN O O
highlights NN O O
the NN O O
importance NN O O
of NN O O
the NN O O
issue NN O O
of NN O O
domain NN O I-OtherScientificTerm
dependence NN O I-OtherScientificTerm
in NN O O
evaluating NN O I-Task
WSD NN O I-Task
programs NN O I-Task
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
an NN O O
analysis NN O O
of NN O O
temporal NN O I-Task
anaphora NN O I-Task
in NN O O
sentences NN O O
which NN O O
contain NN O O
quantification NN O I-OtherScientificTerm
over NN O I-OtherScientificTerm
events NN O I-OtherScientificTerm
, NN O O
within NN O O
the NN O O
framework NN O O
of NN O O
Discourse NN O I-Method
Representation NN O I-Method
Theory NN O I-Method
. NN O O

The NN O O
analysis NN O O
in NN O O
-LRB- NN O O
Partee NN O O
, NN O O
1984 NN O O
-RRB- NN O O
of NN O O
quantified NN O I-Material
sentences NN O I-Material
, NN O O
introduced NN O O
by NN O O
a NN O O
temporal NN O I-OtherScientificTerm
connective NN O I-OtherScientificTerm
, NN O O
gives NN O O
the NN O O
wrong NN O O
truth-conditions NN O O
when NN O O
the NN O O
temporal NN O I-OtherScientificTerm
connective NN O I-OtherScientificTerm
in NN O O
the NN O O
subordinate NN O I-OtherScientificTerm
clause NN O I-OtherScientificTerm
is NN O O
before NN O O
or NN O O
after NN O O
. NN O O

This NN O O
problem NN O I-Generic
has NN O O
been NN O O
previously NN O O
analyzed NN O O
in NN O O
-LRB- NN O O
de NN O O
Swart NN O O
, NN O O
1991 NN O O
-RRB- NN O O
as NN O O
an NN O O
instance NN O O
of NN O O
the NN O O
proportion NN O I-OtherScientificTerm
problem NN O I-OtherScientificTerm
and NN O O
given NN O O
a NN O O
solution NN O I-Generic
from NN O O
a NN O O
Generalized NN O I-Method
Quantifier NN O I-Method
approach NN O I-Method
. NN O O

By NN O O
using NN O O
a NN O O
careful NN O O
distinction NN O O
between NN O O
the NN O O
different NN O O
notions NN O O
of NN O O
reference NN O I-OtherScientificTerm
time NN O I-OtherScientificTerm
based NN O O
on NN O O
-LRB- NN O O
Kamp NN O O
and NN O O
Reyle NN O O
, NN O O
1993 NN O O
-RRB- NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
solution NN O I-Generic
to NN O O
this NN O O
problem NN O I-Generic
, NN O O
within NN O O
the NN O O
framework NN O O
of NN O O
DRT NN O I-Method
. NN O O

We NN O O
show NN O O
some NN O O
applications NN O O
of NN O O
this NN O O
solution NN O I-Generic
to NN O O
additional NN O O
temporal NN O I-Task
anaphora NN O I-Task
phenomena NN O I-Task
in NN O O
quantified NN O I-Material
sentences NN O I-Material
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
explore NN O O
correlation NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
dependency NN O I-OtherScientificTerm
relation NN O I-OtherScientificTerm
paths NN O I-OtherScientificTerm
to NN O O
rank NN O O
candidate NN O O
answers NN O O
in NN O O
answer NN O I-Task
extraction NN O I-Task
. NN O O

Using NN O O
the NN O O
correlation NN O I-Metric
measure NN O I-Metric
, NN O O
we NN O O
compare NN O O
dependency NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
of NN O O
a NN O O
candidate NN O O
answer NN O O
and NN O O
mapped NN O O
question NN O O
phrases NN O O
in NN O O
sentence NN O O
with NN O O
the NN O O
corresponding NN O O
relations NN O O
in NN O O
question NN O O
. NN O O

Different NN O O
from NN O O
previous NN O O
studies NN O O
, NN O O
we NN O O
propose NN O O
an NN O O
approximate NN O I-Method
phrase NN O I-Method
mapping NN O I-Method
algorithm NN O I-Method
and NN O O
incorporate NN O O
the NN O O
mapping NN O I-OtherScientificTerm
score NN O I-OtherScientificTerm
into NN O O
the NN O O
correlation NN O I-Metric
measure NN O I-Metric
. NN O O

The NN O O
correlations NN O I-Generic
are NN O O
further NN O O
incorporated NN O O
into NN O O
a NN O O
Maximum NN O I-Method
Entropy-based NN O I-Method
ranking NN O I-Method
model NN O I-Method
which NN O O
estimates NN O O
path NN O O
weights NN O O
from NN O O
training NN O O
. NN O O

Experimental NN O O
results NN O O
show NN O O
that NN O O
our NN O O
method NN O I-Generic
significantly NN O O
outperforms NN O O
state-of-the-art NN O O
syntactic NN O I-Method
relation-based NN O I-Method
methods NN O I-Method
by NN O O
up NN O O
to NN O O
20 NN O O
% NN O O
in NN O O
MRR NN O I-Metric
. NN O O

-DOCSTART- NN O O

A NN O O
meaningful NN O O
evaluation NN O I-Task
methodology NN O I-Task
can NN O O
advance NN O O
the NN O O
state-of-the-art NN O O
by NN O O
encouraging NN O O
mature NN O O
, NN O O
practical NN O O
applications NN O O
rather NN O O
than NN O O
`` NN O O
toy NN O O
'' NN O O
implementations NN O O
. NN O O

Evaluation NN O I-Task
is NN O O
also NN O O
crucial NN O O
to NN O O
assessing NN O O
competing NN O O
claims NN O O
and NN O O
identifying NN O O
promising NN O O
technical NN O O
approaches NN O I-Generic
. NN O O

While NN O O
work NN O O
in NN O O
speech NN O I-Task
recognition NN O I-Task
-LRB- NN O I-Task
SR NN O I-Task
-RRB- NN O I-Task
has NN O O
a NN O O
history NN O O
of NN O O
evaluation NN O I-Generic
methodologies NN O I-Generic
that NN O O
permit NN O O
comparison NN O O
among NN O O
various NN O O
systems NN O O
, NN O O
until NN O O
recently NN O O
no NN O O
methodology NN O O
existed NN O O
for NN O O
either NN O O
developers NN O O
of NN O O
natural NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
NL NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
interfaces NN O I-OtherScientificTerm
or NN O O
researchers NN O O
in NN O O
speech NN O I-Task
understanding NN O I-Task
-LRB- NN O I-Task
SU NN O I-Task
-RRB- NN O I-Task
to NN O O
evaluate NN O O
and NN O O
compare NN O O
the NN O O
systems NN O O
they NN O O
developed NN O O
. NN O O

Recently NN O O
considerable NN O O
progress NN O O
has NN O O
been NN O O
made NN O O
by NN O O
a NN O O
number NN O O
of NN O O
groups NN O O
involved NN O O
in NN O O
the NN O O
DARPA NN O O
Spoken NN O O
Language NN O O
Systems NN O O
-LRB- NN O O
SLS NN O O
-RRB- NN O O
program NN O O
to NN O O
agree NN O O
on NN O O
a NN O O
methodology NN O I-Generic
for NN O O
comparative NN O O
evaluation NN O I-Task
of NN O I-Task
SLS NN O I-Method
systems NN O I-Method
, NN O O
and NN O O
that NN O O
methodology NN O I-Generic
has NN O O
been NN O O
put NN O O
into NN O O
practice NN O O
several NN O O
times NN O O
in NN O O
comparative NN O O
tests NN O O
of NN O O
several NN O O
SLS NN O I-Method
systems NN O I-Method
. NN O O

These NN O O
evaluations NN O I-Generic
are NN O O
probably NN O O
the NN O O
only NN O O
NL NN O I-Task
evaluations NN O I-Task
other NN O O
than NN O O
the NN O O
series NN O O
of NN O O
Message NN O I-Material
Understanding NN O I-Material
Conferences NN O I-Material
-LRB- NN O O
Sundheim NN O O
, NN O O
1989 NN O O
; NN O O
Sundheim NN O O
, NN O O
1991 NN O O
-RRB- NN O O
to NN O O
have NN O O
been NN O O
developed NN O O
and NN O O
used NN O O
by NN O O
a NN O O
group NN O O
of NN O O
researchers NN O O
at NN O O
different NN O O
sites NN O O
, NN O O
although NN O O
several NN O O
excellent NN O O
workshops NN O O
have NN O O
been NN O O
held NN O O
to NN O O
study NN O O
some NN O O
of NN O O
these NN O O
problems NN O O
-LRB- NN O O
Palmer NN O O
et NN O O
al. NN O O
, NN O O
1989 NN O O
; NN O O
Neal NN O O
et NN O O
al. NN O O
, NN O O
1991 NN O O
-RRB- NN O O
. NN O O

This NN O O
paper NN O O
describes NN O O
a NN O O
practical NN O O
`` NN O I-Metric
black-box NN O I-Metric
'' NN O I-Metric
methodology NN O I-Metric
for NN O O
automatic NN O I-Task
evaluation NN O I-Task
of NN O I-Task
question-answering NN O I-Task
NL NN O I-Task
systems NN O I-Task
. NN O O

While NN O O
each NN O O
new NN O O
application NN O O
domain NN O O
will NN O O
require NN O O
some NN O O
development NN O O
of NN O O
special NN O O
resources NN O O
, NN O O
the NN O O
heart NN O O
of NN O O
the NN O O
methodology NN O I-Generic
is NN O O
domain-independent NN O O
, NN O O
and NN O O
it NN O I-Generic
can NN O O
be NN O O
used NN O O
with NN O O
either NN O O
speech NN O I-OtherScientificTerm
or NN O I-OtherScientificTerm
text NN O I-OtherScientificTerm
input NN O I-OtherScientificTerm
. NN O O

The NN O O
particular NN O O
characteristics NN O O
of NN O O
the NN O O
approach NN O I-Generic
are NN O O
described NN O O
in NN O O
the NN O O
following NN O O
section NN O O
: NN O O
subsequent NN O O
sections NN O O
present NN O O
its NN O O
implementation NN O O
in NN O O
the NN O O
DARPA NN O O
SLS NN O O
community NN O O
, NN O O
and NN O O
some NN O O
problems NN O O
and NN O O
directions NN O O
for NN O O
future NN O O
development NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
present NN O O
a NN O O
novel NN O O
autonomous NN O I-Generic
pipeline NN O I-Generic
to NN O O
build NN O O
a NN O O
personalized NN O I-Method
parametric NN O I-Method
model NN O I-Method
-LRB- NN O I-Method
pose-driven NN O I-Method
avatar NN O I-Method
-RRB- NN O I-Method
using NN O O
a NN O O
single NN O I-OtherScientificTerm
depth NN O I-OtherScientificTerm
sensor NN O I-OtherScientificTerm
. NN O O

Our NN O O
method NN O I-Generic
first NN O O
captures NN O O
a NN O O
few NN O O
high-quality NN O O
scans NN O O
of NN O O
the NN O O
user NN O O
rotating NN O O
herself NN O O
at NN O O
multiple NN O O
poses NN O O
from NN O O
different NN O O
views NN O O
. NN O O

We NN O O
fit NN O O
each NN O O
incomplete NN O O
scan NN O O
using NN O O
template NN O I-Method
fitting NN O I-Method
techniques NN O I-Method
with NN O O
a NN O O
generic NN O O
human NN O I-OtherScientificTerm
template NN O I-OtherScientificTerm
, NN O O
and NN O O
register NN O O
all NN O O
scans NN O O
to NN O O
every NN O O
pose NN O O
using NN O O
global NN O I-OtherScientificTerm
consistency NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
. NN O O

After NN O O
registration NN O O
, NN O O
these NN O O
watertight NN O I-Method
models NN O I-Method
with NN O O
different NN O O
poses NN O O
are NN O O
used NN O O
to NN O O
train NN O O
a NN O O
parametric NN O I-Method
model NN O I-Method
in NN O O
a NN O O
fashion NN O O
similar NN O O
to NN O O
the NN O O
SCAPE NN O I-Method
method NN O I-Method
. NN O O

Once NN O O
the NN O O
parametric NN O I-Method
model NN O I-Method
is NN O O
built NN O O
, NN O O
it NN O I-Generic
can NN O O
be NN O O
used NN O O
as NN O O
an NN O O
anim-itable NN O I-OtherScientificTerm
avatar NN O I-OtherScientificTerm
or NN O O
more NN O O
interestingly NN O O
synthesizing NN O O
dynamic NN O I-Method
3D NN O I-Method
models NN O I-Method
from NN O O
single-view NN O I-Material
depth NN O I-Material
videos NN O I-Material
. NN O O

Experimental NN O O
results NN O O
demonstrate NN O O
the NN O O
effectiveness NN O O
of NN O O
our NN O O
system NN O I-Generic
to NN O O
produce NN O O
dynamic NN O I-Method
models NN O I-Method
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
novel NN O O
algorithm NN O I-Generic
to NN O O
detect/compensate NN O O
on-line NN O I-Task
interference NN O I-Task
effects NN O I-Task
when NN O O
integrating NN O O
Global NN O I-Task
Navigation NN O I-Task
Satellite NN O I-Task
System NN O I-Task
-LRB- NN O I-Task
GNSS NN O I-Task
-RRB- NN O I-Task
and NN O O
Inertial NN O I-Task
Navigation NN O I-Task
System NN O I-Task
-LRB- NN O I-Task
INS NN O I-Task
-RRB- NN O I-Task
. NN O O

The NN O O
GNSS/INS NN O I-Task
coupling NN O I-Task
is NN O O
usually NN O O
performed NN O O
by NN O O
an NN O O
Extended NN O I-Method
Kalman NN O I-Method
Filter NN O I-Method
-LRB- NN O I-Method
EKF NN O I-Method
-RRB- NN O I-Method
which NN O O
yields NN O O
an NN O O
accurate NN O I-Task
and NN O I-Task
robust NN O I-Task
localization NN O I-Task
. NN O O

However NN O O
, NN O O
interference NN O I-OtherScientificTerm
cause NN O O
the NN O O
GNSS NN O I-OtherScientificTerm
measurement NN O I-OtherScientificTerm
noise NN O I-OtherScientificTerm
to NN O O
increase NN O O
unexpectedly NN O O
, NN O O
hence NN O O
degrade NN O O
the NN O O
positioning NN O I-Metric
accuracy NN O I-Metric
. NN O O

In NN O O
this NN O O
context NN O O
, NN O O
our NN O O
contribution NN O O
is NN O O
twofold NN O O
. NN O O

We NN O O
first NN O O
study NN O O
the NN O O
impact NN O O
of NN O O
the NN O O
GNSS NN O O
noise NN O O
inflation NN O O
on NN O O
the NN O O
covariance NN O I-OtherScientificTerm
of NN O O
the NN O O
EKF NN O I-OtherScientificTerm
outputs NN O I-OtherScientificTerm
so NN O O
as NN O O
to NN O O
compute NN O O
a NN O O
least NN O I-Method
square NN O I-Method
estimate NN O I-Method
of NN O O
the NN O O
potential NN O O
variance NN O I-OtherScientificTerm
jumps NN O I-OtherScientificTerm
. NN O O

Then NN O O
, NN O O
this NN O O
estimation NN O I-Generic
is NN O O
used NN O O
in NN O O
a NN O O
Bayesian NN O I-Method
test NN O I-Method
which NN O O
decides NN O O
whether NN O O
interference NN O O
are NN O O
corrupting NN O O
the NN O O
GNSS NN O I-OtherScientificTerm
signal NN O I-OtherScientificTerm
or NN O O
not NN O O
. NN O O

It NN O O
allows NN O O
us NN O O
to NN O O
estimate NN O O
their NN O O
times NN O O
of NN O O
occurrence NN O O
as NN O O
well NN O O
. NN O O

In NN O O
this NN O O
way NN O O
, NN O O
the NN O O
impaired NN O O
measurements NN O O
can NN O O
be NN O O
discarded NN O O
while NN O O
their NN O O
impact NN O O
on NN O O
the NN O O
navigation NN O I-Task
solution NN O I-Task
can NN O O
be NN O O
compensated NN O O
. NN O O

The NN O O
results NN O O
show NN O O
the NN O O
performance NN O O
of NN O O
the NN O O
proposed NN O O
approach NN O I-Generic
on NN O O
simulated NN O I-Material
data NN O I-Material
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
a NN O O
unified NN O I-Method
variational NN O I-Method
formulation NN O I-Method
for NN O O
joint NN O I-Task
motion NN O I-Task
estimation NN O I-Task
and NN O I-Task
segmentation NN O I-Task
with NN O O
explicit NN O I-Method
occlusion NN O I-Method
handling NN O I-Method
. NN O O

This NN O O
is NN O O
done NN O O
by NN O O
a NN O O
multi-label NN O I-Method
representation NN O I-Method
of NN O I-Method
the NN O I-Method
flow NN O I-Method
field NN O I-Method
, NN O O
where NN O O
each NN O O
label NN O O
corresponds NN O O
to NN O O
a NN O O
parametric NN O I-Task
representation NN O I-Task
of NN O I-Task
the NN O I-Task
motion NN O I-Task
. NN O O

We NN O O
use NN O O
a NN O O
convex NN O I-Method
formulation NN O I-Method
of NN O O
the NN O O
multi-label NN O I-Method
Potts NN O I-Method
model NN O I-Method
with NN O O
label NN O O
costs NN O O
and NN O O
show NN O O
that NN O O
the NN O O
asymmetric NN O I-Metric
map-uniqueness NN O I-Metric
criterion NN O I-Metric
can NN O O
be NN O O
integrated NN O O
into NN O O
our NN O O
formulation NN O I-Generic
by NN O O
means NN O O
of NN O O
convex NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
. NN O O

Explicit NN O I-Method
occlusion NN O I-Method
handling NN O I-Method
eliminates NN O O
errors NN O O
otherwise NN O O
created NN O O
by NN O O
the NN O O
regularization NN O I-OtherScientificTerm
. NN O O

As NN O O
occlusions NN O I-OtherScientificTerm
can NN O O
occur NN O O
only NN O O
at NN O O
object NN O I-OtherScientificTerm
boundaries NN O I-OtherScientificTerm
, NN O O
a NN O O
large NN O O
number NN O O
of NN O O
objects NN O O
may NN O O
be NN O O
required NN O O
. NN O O

By NN O O
using NN O O
a NN O O
fast NN O O
primal-dual NN O I-Method
algorithm NN O I-Method
we NN O O
are NN O O
able NN O O
to NN O O
handle NN O O
several NN O O
hundred NN O O
motion NN O I-OtherScientificTerm
segments NN O I-OtherScientificTerm
. NN O O

Results NN O O
are NN O O
shown NN O O
on NN O O
several NN O O
classical NN O I-Task
motion NN O I-Task
segmentation NN O I-Task
and NN O O
optical NN O I-Task
flow NN O I-Task
examples NN O O
. NN O O

-DOCSTART- NN O O

Two NN O O
main NN O O
classes NN O O
of NN O O
approaches NN O I-Generic
have NN O O
been NN O O
studied NN O O
to NN O O
perform NN O O
monocular NN O I-Task
nonrigid NN O I-Task
3D NN O I-Task
reconstruction NN O I-Task
: NN O O
Template-based NN O I-Method
methods NN O I-Method
and NN O O
Non-rigid NN O I-Method
Structure NN O I-Method
from NN O I-Method
Motion NN O I-Method
techniques NN O I-Method
. NN O O

While NN O O
the NN O O
first NN O O
ones NN O I-Generic
have NN O O
been NN O O
applied NN O O
to NN O O
reconstruct NN O O
poorly-textured NN O I-OtherScientificTerm
surfaces NN O I-OtherScientificTerm
, NN O O
they NN O I-Generic
assume NN O O
the NN O O
availability NN O O
of NN O O
a NN O O
3D NN O I-Method
shape NN O I-Method
model NN O I-Method
prior NN O O
to NN O O
reconstruction NN O I-Task
. NN O O

By NN O O
contrast NN O O
, NN O O
the NN O O
second NN O O
ones NN O I-Generic
do NN O O
not NN O O
require NN O O
such NN O O
a NN O O
shape NN O I-OtherScientificTerm
template NN O I-OtherScientificTerm
, NN O O
but NN O O
, NN O O
instead NN O O
, NN O O
rely NN O O
on NN O O
points NN O O
being NN O O
tracked NN O O
throughout NN O O
a NN O O
video NN O I-Material
sequence NN O I-Material
, NN O O
and NN O O
are NN O O
thus NN O O
ill-suited NN O O
to NN O O
handle NN O O
poorly-textured NN O I-OtherScientificTerm
surfaces NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
introduce NN O O
a NN O O
template-free NN O I-Method
approach NN O I-Method
to NN O O
reconstructing NN O O
a NN O O
poorly-textured NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
deformable NN O I-OtherScientificTerm
surface NN O I-OtherScientificTerm
. NN O O

To NN O O
this NN O O
end NN O O
, NN O O
we NN O O
leverage NN O O
surface NN O I-Method
isometry NN O I-Method
and NN O O
formulate NN O O
3D NN O I-Task
reconstruction NN O I-Task
as NN O O
the NN O O
joint NN O I-Task
problem NN O I-Task
of NN O I-Task
non-rigid NN O I-Task
image NN O I-Task
registration NN O I-Task
and NN O I-Task
depth NN O I-Task
estimation NN O I-Task
. NN O O

Our NN O O
experiments NN O O
demonstrate NN O O
that NN O O
our NN O O
approach NN O I-Generic
yields NN O O
much NN O O
more NN O O
accurate NN O O
3D NN O I-Task
reconstructions NN O I-Task
than NN O O
state-of-the-art NN O I-Generic
techniques NN O I-Generic
. NN O O

-DOCSTART- NN O O

Many NN O O
computer NN O I-Task
vision NN O I-Task
applications NN O I-Task
, NN O O
such NN O O
as NN O O
image NN O I-Task
classification NN O I-Task
and NN O O
video NN O I-Task
indexing NN O I-Task
, NN O O
are NN O O
usually NN O O
multi-label NN O I-Task
classification NN O I-Task
problems NN O I-Task
in NN O O
which NN O O
an NN O O
instance NN O O
can NN O O
be NN O O
assigned NN O O
to NN O O
more NN O O
than NN O O
one NN O O
category NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
a NN O O
novel NN O O
multi-label NN O I-Method
classification NN O I-Method
approach NN O I-Method
with NN O O
hypergraph NN O I-OtherScientificTerm
regu-larization NN O I-OtherScientificTerm
that NN O O
addresses NN O O
the NN O O
correlations NN O O
among NN O O
different NN O O
categories NN O O
. NN O O

First NN O O
, NN O O
a NN O O
hypergraph NN O I-OtherScientificTerm
is NN O O
constructed NN O O
to NN O O
capture NN O O
the NN O O
correlations NN O O
among NN O O
different NN O O
categories NN O O
, NN O O
in NN O O
which NN O O
each NN O O
vertex NN O O
represents NN O O
one NN O O
training NN O O
instance NN O O
and NN O O
each NN O O
hyperedge NN O O
for NN O O
one NN O O
category NN O O
contains NN O O
all NN O O
the NN O O
instances NN O O
belonging NN O O
to NN O O
the NN O O
same NN O O
category NN O O
. NN O O

Then NN O O
, NN O O
an NN O O
improved NN O O
SVM NN O I-Method
like NN O I-Method
learning NN O I-Method
system NN O I-Method
incorporating NN O O
the NN O O
hypergraph NN O I-OtherScientificTerm
regularization NN O I-OtherScientificTerm
, NN O O
called NN O O
Rank-HLapSVM NN O I-Method
, NN O O
is NN O O
proposed NN O O
to NN O O
handle NN O O
the NN O O
multi-label NN O I-Task
classification NN O I-Task
problems NN O I-Task
. NN O O

We NN O O
find NN O O
that NN O O
the NN O O
corresponding NN O O
optimization NN O I-Task
problem NN O I-Task
can NN O O
be NN O O
efficiently NN O O
solved NN O O
by NN O O
the NN O O
dual NN O I-Method
coordinate NN O I-Method
descent NN O I-Method
method NN O I-Method
. NN O O

Many NN O O
promising NN O O
experimental NN O O
results NN O O
on NN O O
the NN O O
real NN O I-Generic
datasets NN O I-Generic
including NN O O
ImageCLEF NN O I-Material
and NN O O
Me-diaMill NN O I-Material
demonstrate NN O O
the NN O O
effectiveness NN O O
and NN O O
efficiency NN O O
of NN O O
the NN O O
proposed NN O O
algorithm NN O I-Generic
. NN O O

-DOCSTART- NN O O

We NN O O
derive NN O O
a NN O O
convex NN O I-Task
optimization NN O I-Task
problem NN O I-Task
for NN O O
the NN O O
task NN O O
of NN O O
segmenting NN O I-Task
sequential NN O I-Task
data NN O I-Task
, NN O O
which NN O O
explicitly NN O O
treats NN O O
presence NN O O
of NN O O
outliers NN O I-OtherScientificTerm
. NN O O

We NN O O
describe NN O O
two NN O O
algorithms NN O I-Generic
for NN O O
solving NN O O
this NN O O
problem NN O I-Generic
, NN O O
one NN O O
exact NN O O
and NN O O
one NN O O
a NN O O
top-down NN O O
novel NN O O
approach NN O O
, NN O O
and NN O O
we NN O O
derive NN O O
a NN O O
consistency NN O O
results NN O O
for NN O O
the NN O O
case NN O O
of NN O O
two NN O O
segments NN O O
and NN O O
no NN O O
outliers NN O O
. NN O O

Robustness NN O I-Metric
to NN O O
outliers NN O I-OtherScientificTerm
is NN O O
evaluated NN O O
on NN O O
two NN O O
real-world NN O I-Task
tasks NN O I-Task
related NN O O
to NN O O
speech NN O I-Task
segmentation NN O I-Task
. NN O O

Our NN O O
algorithms NN O I-Generic
outperform NN O O
baseline NN O I-Method
seg-mentation NN O I-Method
algorithms NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
examines NN O O
the NN O O
properties NN O O
of NN O O
feature-based NN O I-OtherScientificTerm
partial NN O I-OtherScientificTerm
descriptions NN O I-OtherScientificTerm
built NN O O
on NN O O
top NN O O
of NN O O
Halliday NN O I-Method
's NN O I-Method
systemic NN O I-Method
networks NN O I-Method
. NN O O

We NN O O
show NN O O
that NN O O
the NN O O
crucial NN O O
operation NN O O
of NN O O
consistency NN O I-Method
checking NN O I-Method
for NN O O
such NN O O
descriptions NN O I-Generic
is NN O O
NP-complete NN O O
, NN O O
and NN O O
therefore NN O O
probably NN O O
intractable NN O O
, NN O O
but NN O O
proceed NN O O
to NN O O
develop NN O O
algorithms NN O O
which NN O O
can NN O O
sometimes NN O O
alleviate NN O O
the NN O O
unpleasant NN O O
consequences NN O O
of NN O O
this NN O O
intractability NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
Yoopick NN O I-Method
, NN O O
a NN O O
combinatorial NN O I-Method
sports NN O I-Method
prediction NN O I-Method
market NN O I-Method
that NN O O
implements NN O O
a NN O O
flexible NN O I-OtherScientificTerm
betting NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
, NN O O
and NN O O
in NN O O
turn NN O O
facilitates NN O O
fine-grained NN O I-Task
probabilistic NN O I-Task
estimation NN O I-Task
of NN O I-Task
outcomes NN O I-Task
. NN O O

-DOCSTART- NN O O

The NN O O
goal NN O O
of NN O O
this NN O O
paper NN O O
is NN O O
to NN O O
discover NN O O
a NN O O
set NN O O
of NN O O
discriminative NN O I-Method
patches NN O I-Method
which NN O O
can NN O O
serve NN O O
as NN O O
a NN O O
fully NN O O
unsupervised NN O I-Task
mid-level NN O I-Task
visual NN O I-Task
representation NN O I-Task
. NN O O

The NN O O
desired NN O O
patches NN O O
need NN O O
to NN O O
satisfy NN O O
two NN O O
requirements NN O O
: NN O O
1 NN O O
-RRB- NN O O
to NN O O
be NN O O
representative NN O O
, NN O O
they NN O O
need NN O O
to NN O O
occur NN O O
frequently NN O O
enough NN O O
in NN O O
the NN O O
visual NN O O
world NN O O
; NN O O
2 NN O O
-RRB- NN O O
to NN O O
be NN O O
discriminative NN O O
, NN O O
they NN O O
need NN O O
to NN O O
be NN O O
different NN O O
enough NN O O
from NN O O
the NN O O
rest NN O O
of NN O O
the NN O O
visual NN O O
world NN O O
. NN O O

The NN O O
patches NN O I-Generic
could NN O O
correspond NN O O
to NN O O
parts NN O O
, NN O O
objects NN O O
, NN O O
'' NN O O
visual NN O O
phrases NN O O
'' NN O O
, NN O O
etc. NN O O
but NN O O
are NN O O
not NN O O
restricted NN O O
to NN O O
be NN O O
any NN O O
one NN O O
of NN O O
them NN O O
. NN O O

We NN O O
pose NN O O
this NN O I-Generic
as NN O O
an NN O O
unsupervised NN O I-Task
discriminative NN O I-Task
clustering NN O I-Task
problem NN O I-Task
on NN O O
a NN O O
huge NN O O
dataset NN O O
of NN O O
image NN O I-Material
patches NN O I-Material
. NN O O

We NN O O
use NN O O
an NN O O
iterative NN O I-Method
procedure NN O I-Method
which NN O O
alternates NN O O
between NN O O
clustering NN O I-Method
and NN O O
training NN O I-Method
discriminative NN O I-Method
classifiers NN O I-Method
, NN O O
while NN O O
applying NN O O
careful NN O O
cross-validation NN O I-Method
at NN O O
each NN O O
step NN O O
to NN O O
prevent NN O O
overfitting NN O I-OtherScientificTerm
. NN O O

The NN O O
paper NN O O
experimentally NN O O
demonstrates NN O O
the NN O O
effectiveness NN O O
of NN O O
discriminative NN O I-Method
patches NN O I-Method
as NN O O
an NN O O
unsupervised NN O I-Task
mid-level NN O I-Task
visual NN O I-Task
representation NN O I-Task
, NN O O
suggesting NN O O
that NN O O
it NN O I-Generic
could NN O O
be NN O O
used NN O O
in NN O O
place NN O O
of NN O O
visual NN O O
words NN O O
for NN O O
many NN O O
tasks NN O O
. NN O O

Furthermore NN O O
, NN O O
discrim-inative NN O I-Method
patches NN O I-Method
can NN O O
also NN O O
be NN O O
used NN O O
in NN O O
a NN O O
supervised NN O I-Task
regime NN O I-Task
, NN O O
such NN O O
as NN O O
scene NN O I-Task
classification NN O I-Task
, NN O O
where NN O O
they NN O I-Generic
demonstrate NN O O
state-of-the-art NN O O
performance NN O O
on NN O O
the NN O O
MIT NN O I-Material
Indoor-67 NN O I-Material
dataset NN O I-Material
. NN O O

-DOCSTART- NN O O

We NN O O
investigate NN O O
the NN O O
utility NN O O
of NN O O
an NN O O
algorithm NN O I-Generic
for NN O O
translation NN O I-Task
lexicon NN O I-Task
acquisition NN O I-Task
-LRB- NN O I-Task
SABLE NN O I-Task
-RRB- NN O I-Task
, NN O O
used NN O O
previously NN O O
on NN O O
a NN O O
very NN O O
large NN O O
corpus NN O O
to NN O O
acquire NN O O
general NN O I-OtherScientificTerm
translation NN O I-OtherScientificTerm
lexicons NN O I-OtherScientificTerm
, NN O O
when NN O O
that NN O O
algorithm NN O I-Generic
is NN O O
applied NN O O
to NN O O
a NN O O
much NN O O
smaller NN O O
corpus NN O O
to NN O O
produce NN O O
candidates NN O O
for NN O O
domain-specific NN O I-OtherScientificTerm
translation NN O I-OtherScientificTerm
lexicons NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
a NN O O
computational NN O I-Method
model NN O I-Method
of NN O O
word NN O I-Task
segmentation NN O I-Task
and NN O O
presents NN O O
simulation NN O O
results NN O O
on NN O O
realistic NN O I-Task
acquisition NN O I-Task
. NN O O

In NN O O
particular NN O O
, NN O O
we NN O O
explore NN O O
the NN O O
capacity NN O O
and NN O O
limitations NN O O
of NN O O
statistical NN O I-Method
learning NN O I-Method
mechanisms NN O I-Method
that NN O O
have NN O O
recently NN O O
gained NN O O
prominence NN O O
in NN O O
cognitive NN O I-Material
psychology NN O I-Material
and NN O O
linguistics NN O I-Material
. NN O O

-DOCSTART- NN O O

In NN O O
the NN O O
model-based NN O I-Method
policy NN O I-Method
search NN O I-Method
approach NN O I-Method
to NN O O
reinforcement NN O I-Task
learning NN O I-Task
-LRB- NN O I-Task
RL NN O I-Task
-RRB- NN O I-Task
, NN O O
policies NN O I-OtherScientificTerm
are NN O O
found NN O O
using NN O O
a NN O O
model NN O O
-LRB- NN O O
or NN O O
`` NN O O
simulator NN O O
'' NN O O
-RRB- NN O O
of NN O O
the NN O O
Markov NN O I-OtherScientificTerm
decision NN O I-OtherScientificTerm
process NN O I-OtherScientificTerm
. NN O O

However NN O O
, NN O O
for NN O O
high-dimensional NN O I-Task
continuous-state NN O I-Task
tasks NN O I-Task
, NN O O
it NN O O
can NN O O
be NN O O
extremely NN O O
difficult NN O O
to NN O O
build NN O O
an NN O O
accurate NN O O
model NN O I-Generic
, NN O O
and NN O O
thus NN O O
often NN O O
the NN O O
algorithm NN O I-Generic
returns NN O O
a NN O O
policy NN O I-OtherScientificTerm
that NN O O
works NN O O
in NN O O
simulation NN O O
but NN O O
not NN O O
in NN O O
real-life NN O O
. NN O O

The NN O O
other NN O O
extreme NN O O
, NN O O
model-free NN O I-Task
RL NN O I-Task
, NN O O
tends NN O O
to NN O O
require NN O O
infeasibly NN O O
large NN O O
numbers NN O O
of NN O O
real-life NN O I-OtherScientificTerm
trials NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
a NN O O
hybrid NN O I-Method
algorithm NN O I-Method
that NN O O
requires NN O O
only NN O O
an NN O O
approximate NN O I-Method
model NN O I-Method
, NN O O
and NN O O
only NN O O
a NN O O
small NN O O
number NN O O
of NN O O
real-life NN O I-OtherScientificTerm
trials NN O I-OtherScientificTerm
. NN O O

The NN O O
key NN O O
idea NN O O
is NN O O
to NN O O
successively NN O O
`` NN O O
ground NN O O
'' NN O O
the NN O O
policy NN O I-Task
evaluations NN O I-Task
using NN O O
real-life NN O I-OtherScientificTerm
trials NN O I-OtherScientificTerm
, NN O O
but NN O O
to NN O O
rely NN O O
on NN O O
the NN O O
approximate NN O I-Method
model NN O I-Method
to NN O O
suggest NN O O
local NN O O
changes NN O O
. NN O O

Our NN O O
theoretical NN O O
results NN O O
show NN O O
that NN O O
this NN O O
algorithm NN O I-Generic
achieves NN O O
near-optimal NN O I-OtherScientificTerm
performance NN O I-OtherScientificTerm
in NN O O
the NN O O
real NN O O
system NN O O
, NN O O
even NN O O
when NN O O
the NN O O
model NN O O
is NN O O
only NN O O
approximate NN O O
. NN O O

Empirical NN O O
results NN O O
also NN O O
demonstrate NN O O
that NN O O
-- NN O O
when NN O O
given NN O O
only NN O O
a NN O O
crude NN O I-Generic
model NN O I-Generic
and NN O O
a NN O O
small NN O O
number NN O O
of NN O O
real-life NN O I-OtherScientificTerm
trials NN O I-OtherScientificTerm
-- NN O O
our NN O O
algorithm NN O I-Generic
can NN O O
obtain NN O O
near-optimal NN O I-OtherScientificTerm
performance NN O I-OtherScientificTerm
in NN O O
the NN O O
real NN O O
system NN O O
. NN O O

-DOCSTART- NN O O

Although NN O O
every NN O O
natural NN O I-Method
language NN O I-Method
system NN O I-Method
needs NN O O
a NN O O
computational NN O I-OtherScientificTerm
lexicon NN O I-OtherScientificTerm
, NN O O
each NN O O
system NN O I-Generic
puts NN O O
different NN O O
amounts NN O O
and NN O O
types NN O O
of NN O O
information NN O O
into NN O O
its NN O O
lexicon NN O O
according NN O O
to NN O O
its NN O O
individual NN O O
needs NN O O
. NN O O

However NN O O
, NN O O
some NN O O
of NN O O
the NN O O
information NN O O
needed NN O O
across NN O O
systems NN O O
is NN O O
shared NN O O
or NN O O
identical NN O O
information NN O O
. NN O O

This NN O O
paper NN O O
presents NN O O
our NN O O
experience NN O O
in NN O O
planning NN O O
and NN O O
building NN O O
COMPLEX NN O I-Method
, NN O O
a NN O O
computational NN O I-OtherScientificTerm
lexicon NN O I-OtherScientificTerm
designed NN O O
to NN O O
be NN O O
a NN O O
repository NN O O
of NN O O
shared NN O I-OtherScientificTerm
lexical NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
for NN O O
use NN O O
by NN O O
Natural NN O I-Task
Language NN O I-Task
Processing NN O I-Task
-LRB- NN O I-Task
NLP NN O I-Task
-RRB- NN O I-Task
systems NN O I-Task
. NN O O

We NN O O
have NN O O
drawn NN O O
primarily NN O O
on NN O O
explicit NN O O
and NN O O
implicit NN O O
information NN O O
from NN O O
machine-readable NN O I-Material
dictionaries NN O I-Material
-LRB- NN O I-Material
MRD NN O I-Material
's NN O I-Material
-RRB- NN O I-Material
to NN O O
create NN O O
a NN O O
broad NN O O
coverage NN O O
lexicon NN O O
. NN O O

-DOCSTART- NN O O

Sentence NN O I-Task
planning NN O I-Task
is NN O O
a NN O O
set NN O O
of NN O O
inter-related NN O O
but NN O O
distinct NN O O
tasks NN O I-Generic
, NN O O
one NN O O
of NN O O
which NN O O
is NN O O
sentence NN O I-Task
scoping NN O I-Task
, NN O O
i.e. NN O O
the NN O O
choice NN O O
of NN O O
syntactic NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
for NN O O
elementary NN O O
speech NN O I-OtherScientificTerm
acts NN O I-OtherScientificTerm
and NN O O
the NN O O
decision NN O O
of NN O O
how NN O O
to NN O O
combine NN O O
them NN O O
into NN O O
one NN O O
or NN O O
more NN O O
sentences NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
SPoT NN O I-Method
, NN O O
a NN O O
sentence NN O I-Method
planner NN O I-Method
, NN O O
and NN O O
a NN O O
new NN O O
methodology NN O I-Generic
for NN O O
automatically NN O O
training NN O O
SPoT NN O I-Method
on NN O O
the NN O O
basis NN O O
of NN O O
feedback NN O O
provided NN O O
by NN O O
human NN O O
judges NN O O
. NN O O

We NN O O
reconceptualize NN O O
the NN O O
task NN O O
into NN O O
two NN O O
distinct NN O O
phases NN O O
. NN O O

First NN O O
, NN O O
a NN O O
very NN O O
simple NN O O
, NN O O
randomized NN O I-Method
sentence-plan-generator NN O I-Method
-LRB- NN O I-Method
SPG NN O I-Method
-RRB- NN O I-Method
generates NN O O
a NN O O
potentially NN O O
large NN O O
list NN O O
of NN O O
possible NN O O
sentence NN O I-OtherScientificTerm
plans NN O I-OtherScientificTerm
for NN O O
a NN O O
given NN O O
text-plan NN O I-OtherScientificTerm
input NN O I-OtherScientificTerm
. NN O O

Second NN O O
, NN O O
the NN O O
sentence-plan-ranker NN O I-Method
-LRB- NN O I-Method
SPR NN O I-Method
-RRB- NN O I-Method
ranks NN O O
the NN O O
list NN O O
of NN O O
output NN O O
sentence NN O I-OtherScientificTerm
plans NN O I-OtherScientificTerm
, NN O O
and NN O O
then NN O O
selects NN O O
the NN O O
top-ranked NN O O
plan NN O O
. NN O O

The NN O O
SPR NN O I-Method
uses NN O O
ranking NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
automatically NN O O
learned NN O O
from NN O O
training NN O O
data NN O O
. NN O O

We NN O O
show NN O O
that NN O O
the NN O O
trained NN O O
SPR NN O I-Method
learns NN O O
to NN O O
select NN O O
a NN O O
sentence NN O I-OtherScientificTerm
plan NN O I-OtherScientificTerm
whose NN O O
rating NN O O
on NN O O
average NN O O
is NN O O
only NN O O
5 NN O O
% NN O O
worse NN O O
than NN O O
the NN O O
top NN O I-OtherScientificTerm
human-ranked NN O I-OtherScientificTerm
sentence NN O I-OtherScientificTerm
plan NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

We NN O O
discuss NN O O
maximum NN O I-Method
a NN O I-Method
posteriori NN O I-Method
estimation NN O I-Method
of NN O O
continuous NN O I-Method
density NN O I-Method
hidden NN O I-Method
Markov NN O I-Method
models NN O I-Method
-LRB- NN O I-Method
CDHMM NN O I-Method
-RRB- NN O I-Method
. NN O O

The NN O O
classical NN O O
MLE NN O I-Method
reestimation NN O I-Method
algorithms NN O I-Method
, NN O O
namely NN O O
the NN O O
forward-backward NN O I-Method
algorithm NN O I-Method
and NN O O
the NN O O
segmental NN O I-Method
k-means NN O I-Method
algorithm NN O I-Method
, NN O O
are NN O O
expanded NN O O
and NN O O
reestimation NN O I-OtherScientificTerm
formulas NN O I-OtherScientificTerm
are NN O O
given NN O O
for NN O O
HMM NN O I-Method
with NN O I-Method
Gaussian NN O I-Method
mixture NN O I-Method
observation NN O I-Method
densities NN O I-Method
. NN O O

Because NN O O
of NN O O
its NN O O
adaptive NN O O
nature NN O O
, NN O O
Bayesian NN O I-Method
learning NN O I-Method
serves NN O O
as NN O O
a NN O O
unified NN O O
approach NN O O
for NN O O
the NN O O
following NN O O
four NN O O
speech NN O I-Task
recognition NN O I-Task
applications NN O I-Task
, NN O O
namely NN O O
parameter NN O I-Task
smoothing NN O I-Task
, NN O O
speaker NN O I-Task
adaptation NN O I-Task
, NN O O
speaker NN O I-Task
group NN O I-Task
modeling NN O I-Task
and NN O O
corrective NN O I-Task
training NN O I-Task
. NN O O

New NN O O
experimental NN O O
results NN O O
on NN O O
all NN O O
four NN O O
applications NN O I-Generic
are NN O O
provided NN O O
to NN O O
show NN O O
the NN O O
effectiveness NN O O
of NN O O
the NN O O
MAP NN O I-Method
estimation NN O I-Method
approach NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
a NN O O
characters-based NN O I-Method
Chinese NN O I-Method
collocation NN O I-Method
system NN O I-Method
and NN O O
discusses NN O O
the NN O O
advantages NN O O
of NN O O
it NN O I-Generic
over NN O O
a NN O O
traditional NN O O
word-based NN O I-Method
system NN O I-Method
. NN O O

Since NN O O
wordbreaks NN O O
are NN O O
not NN O O
conventionally NN O O
marked NN O O
in NN O O
Chinese NN O I-Material
text NN O I-Material
corpora NN O I-Material
, NN O O
a NN O O
character-based NN O I-Method
collocation NN O I-Method
system NN O I-Method
has NN O O
the NN O O
dual NN O O
advantages NN O O
of NN O O
avoiding NN O I-Task
pre-processing NN O I-Task
distortion NN O I-Task
and NN O O
directly NN O O
accessing NN O I-Task
sub-lexical NN O I-Task
information NN O I-Task
. NN O O

Furthermore NN O O
, NN O O
word-based NN O I-OtherScientificTerm
collocational NN O I-OtherScientificTerm
properties NN O I-OtherScientificTerm
can NN O O
be NN O O
obtained NN O O
through NN O O
an NN O O
auxiliary NN O I-Method
module NN O I-Method
of NN O I-Method
automatic NN O I-Method
segmentation NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
a NN O O
method NN O I-Generic
for NN O O
utterance NN O I-Task
classification NN O I-Task
that NN O O
does NN O O
not NN O O
require NN O O
manual NN O I-Material
transcription NN O I-Material
of NN O O
training NN O O
data NN O O
. NN O O

The NN O O
method NN O I-Generic
combines NN O O
domain NN O I-Method
independent NN O I-Method
acoustic NN O I-Method
models NN O I-Method
with NN O O
off-the-shelf NN O O
classifiers NN O I-Method
to NN O O
give NN O O
utterance NN O I-Task
classification NN O I-Task
performance NN O O
that NN O O
is NN O O
surprisingly NN O O
close NN O O
to NN O O
what NN O O
can NN O O
be NN O O
achieved NN O O
using NN O O
conventional NN O O
word-trigram NN O I-Method
recognition NN O I-Method
requiring NN O O
manual NN O I-Material
transcription NN O I-Material
. NN O O

In NN O O
our NN O O
method NN O I-Generic
, NN O O
unsupervised NN O I-Method
training NN O I-Method
is NN O O
first NN O O
used NN O O
to NN O O
train NN O O
a NN O O
phone NN O I-Method
n-gram NN O I-Method
model NN O I-Method
for NN O O
a NN O O
particular NN O O
domain NN O I-Generic
; NN O O
the NN O O
output NN O O
of NN O O
recognition NN O O
with NN O O
this NN O O
model NN O I-Generic
is NN O O
then NN O O
passed NN O O
to NN O O
a NN O O
phone-string NN O I-Method
classifier NN O I-Method
. NN O O

The NN O O
classification NN O I-Metric
accuracy NN O I-Metric
of NN O O
the NN O O
method NN O I-Generic
is NN O O
evaluated NN O O
on NN O O
three NN O O
different NN O O
spoken NN O I-Material
language NN O I-Material
system NN O I-Material
domains NN O I-Material
. NN O O

-DOCSTART- NN O O

The NN O O
Interval NN O I-Method
Algebra NN O I-Method
-LRB- NN O I-Method
IA NN O I-Method
-RRB- NN O I-Method
and NN O O
a NN O O
subset NN O O
of NN O O
the NN O O
Region NN O I-Method
Connection NN O I-Method
Calculus NN O I-Method
-LRB- NN O I-Method
RCC NN O I-Method
-RRB- NN O I-Method
, NN O O
namely NN O O
RCC-8 NN O I-Method
, NN O O
are NN O O
the NN O O
dominant NN O O
Artificial NN O I-Method
Intelligence NN O I-Method
approaches NN O I-Method
for NN O O
representing NN O I-Task
and NN O I-Task
reasoning NN O I-Task
about NN O I-Task
qualitative NN O I-OtherScientificTerm
temporal NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
topological NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
respectively NN O O
. NN O O

Such NN O O
qualitative NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
can NN O O
be NN O O
formulated NN O O
as NN O O
a NN O O
Qualitative NN O I-Method
Constraint NN O I-Method
Network NN O I-Method
-LRB- NN O I-Method
QCN NN O I-Method
-RRB- NN O I-Method
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
focus NN O O
on NN O O
the NN O O
minimal NN O I-Task
labeling NN O I-Task
problem NN O I-Task
-LRB- NN O I-Task
MLP NN O I-Task
-RRB- NN O I-Task
and NN O O
we NN O O
propose NN O O
an NN O O
algorithm NN O I-Generic
to NN O O
efficiently NN O O
derive NN O O
all NN O O
the NN O O
feasible NN O O
base NN O O
relations NN O O
of NN O O
a NN O O
QCN NN O I-Method
. NN O O

Our NN O O
algorithm NN O I-Generic
considers NN O O
chordal NN O I-Method
QCNs NN O I-Method
and NN O O
a NN O O
new NN O O
form NN O O
of NN O O
partial NN O I-OtherScientificTerm
consistency NN O I-OtherScientificTerm
which NN O O
we NN O O
define NN O O
as NN O O
◆ NN O I-OtherScientificTerm
G-consistency NN O I-OtherScientificTerm
. NN O O

Further NN O O
, NN O O
the NN O O
proposed NN O O
algorithm NN O I-Generic
uses NN O O
tractable NN O O
subclasses NN O O
of NN O O
relations NN O O
having NN O O
a NN O O
specific NN O O
patchwork NN O I-OtherScientificTerm
property NN O I-OtherScientificTerm
for NN O O
which-consistency NN O O
implies NN O O
the NN O O
consistency NN O O
of NN O O
the NN O O
input NN O O
QCN NN O I-Method
. NN O O

Experi-mentations NN O O
with NN O O
QCNs NN O I-Method
of NN O I-Method
IA NN O I-Method
and NN O I-Method
RCC-8 NN O I-Method
show NN O O
the NN O O
importance NN O O
and NN O O
efficiency NN O O
of NN O O
this NN O O
new NN O O
approach NN O I-Generic
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
a NN O O
morphological NN O I-Method
component NN O I-Method
with NN O O
a NN O O
limited NN O O
capability NN O O
to NN O O
automatically NN O O
interpret NN O O
-LRB- NN O O
and NN O O
generate NN O O
-RRB- NN O O
derived NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
is NN O O
presented NN O O
. NN O O

The NN O O
system NN O I-Generic
combines NN O O
an NN O O
extended NN O O
two-level NN O I-OtherScientificTerm
morphology NN O I-OtherScientificTerm
-LSB- NN O O
Trost NN O O
, NN O O
1991a NN O O
; NN O O
Trost NN O O
, NN O O
1991b NN O O
-RSB- NN O O
with NN O O
a NN O O
feature-based NN O I-Method
word NN O I-Method
grammar NN O I-Method
building NN O O
on NN O O
a NN O O
hierarchical NN O I-OtherScientificTerm
lexicon NN O I-OtherScientificTerm
. NN O O

Polymorphemic NN O I-OtherScientificTerm
stems NN O I-OtherScientificTerm
not NN O O
explicitly NN O O
stored NN O O
in NN O O
the NN O O
lexicon NN O O
are NN O O
given NN O O
a NN O O
compositional NN O I-OtherScientificTerm
interpretation NN O I-OtherScientificTerm
. NN O O

That NN O O
way NN O O
the NN O O
system NN O I-Generic
allows NN O O
to NN O O
minimize NN O O
redundancy NN O O
in NN O O
the NN O O
lexicon NN O O
because NN O O
derived NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
that NN O O
are NN O O
transparent NN O O
need NN O O
not NN O O
to NN O O
be NN O O
stored NN O O
explicitly NN O O
. NN O O

Also NN O O
, NN O O
words NN O I-OtherScientificTerm
formed NN O I-OtherScientificTerm
ad-hoc NN O I-OtherScientificTerm
can NN O O
be NN O O
recognized NN O O
correctly NN O O
. NN O O

The NN O O
system NN O I-Generic
is NN O O
implemented NN O O
in NN O O
CommonLisp NN O I-Method
and NN O O
has NN O O
been NN O O
tested NN O O
on NN O O
examples NN O O
from NN O O
German NN O I-Material
derivation NN O I-Material
. NN O O

-DOCSTART- NN O O

Lyric-based NN O I-Task
song NN O I-Task
sentiment NN O I-Task
classification NN O I-Task
seeks NN O O
to NN O O
assign NN O O
songs NN O O
appropriate NN O O
sentiment NN O O
labels NN O O
such NN O O
as NN O O
light-hearted NN O O
heavy-hearted NN O O
. NN O O

Four NN O O
problems NN O O
render NN O O
vector NN O I-Method
space NN O I-Method
model NN O I-Method
-LRB- NN O I-Method
VSM NN O I-Method
-RRB- NN O I-Method
- NN O I-Method
based NN O I-Method
text NN O I-Method
classification NN O I-Method
approach NN O I-Method
ineffective NN O O
: NN O O
1 NN O O
-RRB- NN O O
Many NN O O
words NN O O
within NN O O
song NN O I-Material
lyrics NN O I-Material
actually NN O O
contribute NN O O
little NN O O
to NN O O
sentiment NN O I-OtherScientificTerm
; NN O O
2 NN O O
-RRB- NN O O
Nouns NN O O
and NN O O
verbs NN O O
used NN O O
to NN O O
express NN O O
sentiment NN O O
are NN O O
ambiguous NN O O
; NN O O
3 NN O O
-RRB- NN O O
Negations NN O I-OtherScientificTerm
and NN O O
modifiers NN O I-OtherScientificTerm
around NN O O
the NN O O
sentiment NN O I-OtherScientificTerm
keywords NN O I-OtherScientificTerm
make NN O O
particular NN O O
contributions NN O O
to NN O O
sentiment NN O I-OtherScientificTerm
; NN O O
4 NN O O
-RRB- NN O O
Song NN O I-Material
lyric NN O I-Material
is NN O O
usually NN O O
very NN O O
short NN O O
. NN O O

To NN O O
address NN O O
these NN O O
problems NN O O
, NN O O
the NN O O
sentiment NN O I-Method
vector NN O I-Method
space NN O I-Method
model NN O I-Method
-LRB- NN O I-Method
s-VSM NN O I-Method
-RRB- NN O I-Method
is NN O O
proposed NN O O
to NN O O
represent NN O O
song NN O I-Material
lyric NN O I-Material
document NN O I-Material
. NN O O

The NN O O
preliminary NN O O
experiments NN O O
prove NN O O
that NN O O
the NN O O
s-VSM NN O I-Method
model NN O I-Method
outperforms NN O O
the NN O O
VSM NN O I-Method
model NN O I-Method
in NN O O
the NN O O
lyric-based NN O I-Task
song NN O I-Task
sentiment NN O I-Task
classification NN O I-Task
task NN O I-Task
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
an NN O O
efficient NN O O
algorithm NN O I-Generic
for NN O O
the NN O O
redundancy NN O I-Task
elimination NN O I-Task
problem NN O I-Task
: NN O O
Given NN O O
an NN O O
underspecified NN O I-Method
semantic NN O I-Method
representation NN O I-Method
-LRB- NN O I-Method
USR NN O I-Method
-RRB- NN O I-Method
of NN O O
a NN O O
scope NN O I-OtherScientificTerm
ambiguity NN O I-OtherScientificTerm
, NN O O
compute NN O O
an NN O O
USR NN O I-Method
with NN O O
fewer NN O O
mutually NN O O
equivalent NN O I-Material
readings NN O I-Material
. NN O O

The NN O O
algorithm NN O I-Generic
operates NN O O
on NN O O
underspecified NN O I-Method
chart NN O I-Method
representations NN O I-Method
which NN O O
are NN O O
derived NN O O
from NN O O
dominance NN O I-OtherScientificTerm
graphs NN O I-OtherScientificTerm
; NN O O
it NN O I-Generic
can NN O O
be NN O O
applied NN O O
to NN O O
the NN O O
USRs NN O I-Method
computed NN O O
by NN O O
large-scale NN O I-Method
grammars NN O I-Method
. NN O O

We NN O O
evaluate NN O O
the NN O O
algorithm NN O I-Generic
on NN O O
a NN O O
corpus NN O O
, NN O O
and NN O O
show NN O O
that NN O O
it NN O I-Generic
reduces NN O O
the NN O O
degree NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
ambiguity NN O I-OtherScientificTerm
significantly NN O O
while NN O O
taking NN O O
negligible NN O O
runtime NN O O
. NN O O

-DOCSTART- NN O O

Currently NN O O
several NN O O
grammatical NN O I-Method
formalisms NN O I-Method
converge NN O O
towards NN O O
being NN O O
declarative NN O O
and NN O O
towards NN O O
utilizing NN O O
context-free NN O I-Method
phrase-structure NN O I-Method
grammar NN O I-Method
as NN O O
a NN O O
backbone NN O O
, NN O O
e.g. NN O O
LFG NN O I-OtherScientificTerm
and NN O O
PATR-II NN O I-OtherScientificTerm
. NN O O

Typically NN O O
the NN O O
processing NN O O
of NN O O
these NN O O
formalisms NN O I-Generic
is NN O O
organized NN O O
within NN O O
a NN O O
chart-parsing NN O I-Method
framework NN O I-Method
. NN O O

The NN O O
declarative NN O O
character NN O O
of NN O O
the NN O O
formalisms NN O I-Generic
makes NN O O
it NN O O
important NN O O
to NN O O
decide NN O O
upon NN O O
an NN O O
overall NN O O
optimal NN O I-Method
control NN O I-Method
strategy NN O I-Method
on NN O O
the NN O O
part NN O O
of NN O O
the NN O O
processor NN O O
. NN O O

In NN O O
particular NN O O
, NN O O
this NN O O
brings NN O O
the NN O O
rule-invocation NN O I-Method
strategy NN O I-Method
into NN O O
critical NN O O
focus NN O O
: NN O O
to NN O O
gain NN O O
maximal NN O O
processing NN O O
efficiency NN O O
, NN O O
one NN O O
has NN O O
to NN O O
determine NN O O
the NN O O
best NN O O
way NN O O
of NN O O
putting NN O O
the NN O O
rules NN O I-OtherScientificTerm
to NN O O
use NN O O
. NN O O

The NN O O
aim NN O O
of NN O O
this NN O O
paper NN O O
is NN O O
to NN O O
provide NN O O
a NN O O
survey NN O O
and NN O O
a NN O O
practical NN O O
comparison NN O O
of NN O O
fundamental NN O O
rule-invocation NN O I-Method
strategies NN O I-Method
within NN O O
context-free NN O I-Method
chart NN O I-Method
parsing NN O I-Method
. NN O O

-DOCSTART- NN O O

Terminology NN O I-Task
structuring NN O I-Task
has NN O O
been NN O O
the NN O O
subject NN O O
of NN O O
much NN O O
work NN O O
in NN O O
the NN O O
context NN O O
of NN O O
terms NN O O
extracted NN O O
from NN O O
corpora NN O O
: NN O O
given NN O O
a NN O O
set NN O O
of NN O O
terms NN O O
, NN O O
obtained NN O O
from NN O O
an NN O O
existing NN O O
resource NN O O
or NN O O
extracted NN O O
from NN O O
a NN O O
corpus NN O I-Material
, NN O O
identifying NN O O
hierarchical NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
or NN O I-OtherScientificTerm
other NN O I-OtherScientificTerm
types NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
between NN O O
these NN O O
terms NN O O
. NN O O

The NN O O
present NN O O
paper NN O O
focusses NN O O
on NN O O
terminology NN O I-Task
structuring NN O I-Task
by NN O O
lexical NN O I-Method
methods NN O I-Method
, NN O O
which NN O O
match NN O O
terms NN O O
on NN O O
the NN O O
basis NN O O
on NN O O
their NN O O
content NN O O
words NN O O
, NN O O
taking NN O O
morphological NN O I-OtherScientificTerm
variants NN O I-OtherScientificTerm
into NN O O
account NN O O
. NN O O

Experiments NN O O
are NN O O
done NN O O
on NN O O
a NN O O
` NN O O
flat NN O O
' NN O O
list NN O O
of NN O O
terms NN O O
obtained NN O O
from NN O O
an NN O O
originally NN O O
hierarchically-structured NN O I-Material
terminology NN O I-Material
: NN O O
the NN O O
French NN O O
version NN O O
of NN O O
the NN O O
US NN O I-Material
National NN O I-Material
Library NN O I-Material
of NN O I-Material
Medicine NN O I-Material
MeSH NN O I-Material
thesaurus NN O I-Material
. NN O O

We NN O O
compare NN O O
the NN O O
lexically-induced NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
with NN O O
the NN O O
original NN O O
MeSH NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
: NN O O
after NN O O
a NN O O
quantitative NN O O
evaluation NN O O
of NN O O
their NN O O
congruence NN O O
through NN O O
recall NN O I-Metric
and NN O I-Metric
precision NN O I-Metric
metrics NN O I-Metric
, NN O O
we NN O O
perform NN O O
a NN O O
qualitative NN O O
, NN O O
human NN O O
analysis NN O O
ofthe NN O O
` NN O O
new NN O O
' NN O O
relations NN O O
not NN O O
present NN O O
in NN O O
the NN O O
MeSH NN O I-OtherScientificTerm
. NN O O

This NN O O
analysis NN O O
shows NN O O
, NN O O
on NN O O
the NN O O
one NN O O
hand NN O O
, NN O O
the NN O O
limits NN O O
of NN O O
the NN O O
lexical NN O I-Method
structuring NN O I-Method
method NN O I-Method
. NN O O

On NN O O
the NN O O
other NN O O
hand NN O O
, NN O O
it NN O O
also NN O O
reveals NN O O
some NN O O
specific NN O O
structuring NN O O
choices NN O O
and NN O O
naming NN O O
conventions NN O O
made NN O O
by NN O O
the NN O O
MeSH NN O I-OtherScientificTerm
designers NN O O
, NN O O
and NN O O
emphasizes NN O O
ontological NN O O
commitments NN O O
that NN O O
can NN O O
not NN O O
be NN O O
left NN O O
to NN O O
automatic NN O I-Task
structuring NN O I-Task
. NN O O

-DOCSTART- NN O O

In NN O O
order NN O O
to NN O O
boost NN O O
the NN O O
translation NN O O
quality NN O O
of NN O O
EBMT NN O I-Method
based NN O O
on NN O O
a NN O O
small-sized NN O I-Material
bilingual NN O I-Material
corpus NN O I-Material
, NN O O
we NN O O
use NN O O
an NN O O
out-of-domain NN O I-Material
bilingual NN O I-Material
corpus NN O I-Material
and NN O O
, NN O O
in NN O O
addition NN O O
, NN O O
the NN O O
language NN O I-Method
model NN O I-Method
of NN O O
an NN O O
in-domain NN O I-Material
monolingual NN O I-Material
corpus NN O I-Material
. NN O O

We NN O O
conducted NN O O
experiments NN O O
with NN O O
an NN O O
EBMT NN O I-Method
system NN O I-Method
. NN O O

The NN O O
two NN O O
evaluation NN O I-Metric
measures NN O I-Metric
of NN O O
the NN O O
BLEU NN O I-Metric
score NN O I-Metric
and NN O O
the NN O O
NIST NN O I-Metric
score NN O I-Metric
demonstrated NN O O
the NN O O
effect NN O O
of NN O O
using NN O O
an NN O O
out-of-domain NN O I-Material
bilingual NN O I-Material
corpus NN O I-Material
and NN O O
the NN O O
possibility NN O O
of NN O O
using NN O O
the NN O O
language NN O I-Method
model NN O I-Method
. NN O O

-DOCSTART- NN O O

Diagrams NN O I-Method
are NN O O
common NN O O
tools NN O O
for NN O O
representing NN O O
complex NN O I-OtherScientificTerm
concepts NN O I-OtherScientificTerm
, NN O O
relationships NN O I-OtherScientificTerm
and NN O O
events NN O I-OtherScientificTerm
, NN O O
often NN O O
when NN O O
it NN O O
would NN O O
be NN O O
difficult NN O O
to NN O O
portray NN O O
the NN O O
same NN O O
information NN O O
with NN O O
natural NN O I-Material
images NN O I-Material
. NN O O

Understanding NN O I-Task
natural NN O I-Material
images NN O I-Material
has NN O O
been NN O O
extensively NN O O
studied NN O O
in NN O O
computer NN O I-Task
vision NN O I-Task
, NN O O
while NN O O
diagram NN O I-Task
understanding NN O I-Task
has NN O O
received NN O O
little NN O O
attention NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
study NN O O
the NN O O
problem NN O O
of NN O O
diagram NN O I-Task
interpretation NN O I-Task
and NN O I-Task
reasoning NN O I-Task
, NN O O
the NN O O
challenging NN O O
task NN O I-Generic
of NN O O
identifying NN O O
the NN O O
structure NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
a NN O I-OtherScientificTerm
diagram NN O I-OtherScientificTerm
and NN O O
the NN O O
semantics NN O O
of NN O O
its NN O O
constituents NN O O
and NN O O
their NN O O
relationships NN O O
. NN O O

We NN O O
introduce NN O O
Diagram NN O I-Method
Parse NN O I-Method
Graphs NN O I-Method
-LRB- NN O I-Method
DPG NN O I-Method
-RRB- NN O I-Method
as NN O O
our NN O O
representation NN O O
to NN O O
model NN O O
the NN O O
structure NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
diagrams NN O I-OtherScientificTerm
. NN O O

We NN O O
define NN O O
syntactic NN O I-Task
parsing NN O I-Task
of NN O I-Task
diagrams NN O I-Task
as NN O O
learning NN O O
to NN O O
infer NN O O
DPGs NN O I-Method
for NN O O
diagrams NN O O
and NN O O
study NN O O
semantic NN O I-Task
interpretation NN O I-Task
and NN O I-Task
reasoning NN O I-Task
of NN O I-Task
diagrams NN O I-Task
in NN O O
the NN O O
context NN O O
of NN O O
diagram NN O I-Task
question NN O I-Task
answering NN O I-Task
. NN O O

We NN O O
devise NN O O
an NN O O
LSTM-based NN O I-Method
method NN O I-Method
for NN O O
syntactic NN O I-Task
parsing NN O I-Task
of NN O I-Task
diagrams NN O I-Task
and NN O O
introduce NN O O
a NN O O
DPG-based NN O I-Method
attention NN O I-Method
model NN O I-Method
for NN O O
diagram NN O I-Task
question NN O I-Task
answering NN O I-Task
. NN O O

We NN O O
compile NN O O
a NN O O
new NN O O
dataset NN O I-Generic
of NN O O
diagrams NN O I-Material
with NN O O
exhaustive NN O O
annotations NN O O
of NN O O
constituents NN O O
and NN O O
relationships NN O O
for NN O O
over NN O O
5,000 NN O O
diagrams NN O O
and NN O O
15,000 NN O O
questions NN O O
and NN O O
answers NN O O
. NN O O

Our NN O O
results NN O O
show NN O O
the NN O O
significance NN O O
of NN O O
our NN O O
models NN O I-Generic
for NN O O
syntactic NN O I-Task
parsing NN O I-Task
and NN O I-Task
question NN O I-Task
answering NN O I-Task
in NN O I-Task
diagrams NN O I-Task
using NN O O
DPGs NN O I-Method
. NN O O

-DOCSTART- NN O O

Detecting NN O I-Task
fine-grained NN O I-Task
subtle NN O I-Task
changes NN O I-Task
among NN O I-Task
a NN O I-Task
scene NN O I-Task
is NN O O
critically NN O O
important NN O O
in NN O O
practice NN O O
. NN O O

Previous NN O O
change NN O I-Method
detection NN O I-Method
methods NN O I-Method
, NN O O
focusing NN O O
on NN O O
detecting NN O I-Task
large-scale NN O I-Task
significant NN O I-Task
changes NN O I-Task
, NN O O
can NN O O
not NN O O
do NN O O
this NN O O
well NN O O
. NN O O

This NN O O
paper NN O O
proposes NN O O
a NN O O
feasible NN O O
end-to-end NN O I-Method
approach NN O I-Method
to NN O O
this NN O O
challenging NN O O
problem NN O I-Generic
. NN O O

We NN O O
start NN O O
from NN O O
active NN O I-Method
camera NN O I-Method
relocation NN O I-Method
that NN O O
quickly NN O O
relocates NN O O
camera NN O O
to NN O O
nearly NN O O
the NN O O
same NN O O
pose NN O O
and NN O O
position NN O O
of NN O O
the NN O O
last NN O O
time NN O O
observation NN O O
. NN O O

To NN O O
guarantee NN O O
detection NN O I-Metric
sensitivity NN O I-Metric
and NN O O
accuracy NN O I-Metric
of NN O O
minute NN O O
changes NN O O
, NN O O
in NN O O
an NN O O
observation NN O O
, NN O O
we NN O O
capture NN O O
a NN O O
group NN O O
of NN O O
images NN O O
under NN O O
multiple NN O O
illuminations NN O I-OtherScientificTerm
, NN O O
which NN O O
need NN O O
only NN O O
to NN O O
be NN O O
roughly NN O O
aligned NN O O
to NN O O
the NN O O
last NN O O
time NN O O
lighting NN O O
conditions NN O O
. NN O O

Given NN O O
two NN O O
times NN O O
observations NN O O
, NN O O
we NN O O
formulate NN O O
fine-grained NN O I-Task
change NN O I-Task
detection NN O I-Task
as NN O O
a NN O O
joint NN O I-Task
optimization NN O I-Task
problem NN O I-Task
of NN O O
three NN O O
related NN O O
factors NN O I-Generic
, NN O O
i.e. NN O O
, NN O O
normal-aware NN O I-OtherScientificTerm
lighting NN O I-OtherScientificTerm
difference NN O I-OtherScientificTerm
, NN O O
camera NN O I-OtherScientificTerm
geometry NN O I-OtherScientificTerm
correction NN O I-OtherScientificTerm
flow NN O I-OtherScientificTerm
, NN O O
and NN O O
real NN O I-OtherScientificTerm
scene NN O I-OtherScientificTerm
change NN O I-OtherScientificTerm
mask NN O I-OtherScientificTerm
. NN O O

We NN O O
solve NN O O
the NN O O
three NN O O
factors NN O I-Generic
in NN O O
a NN O O
coarse-to-fine NN O I-OtherScientificTerm
manner NN O I-OtherScientificTerm
and NN O O
achieve NN O O
reliable NN O O
change NN O I-OtherScientificTerm
decision NN O I-OtherScientificTerm
by NN O O
rank NN O I-Method
minimization NN O I-Method
. NN O O

We NN O O
build NN O O
three NN O O
real-world NN O I-Material
datasets NN O I-Material
to NN O O
benchmark NN O O
fine-grained NN O I-Task
change NN O I-Task
detection NN O I-Task
of NN O I-Task
misaligned NN O I-Task
scenes NN O I-Task
under NN O O
varied NN O I-OtherScientificTerm
multiple NN O I-OtherScientificTerm
lighting NN O I-OtherScientificTerm
conditions NN O I-OtherScientificTerm
. NN O O

Extensive NN O O
experiments NN O O
show NN O O
the NN O O
superior NN O O
performance NN O O
of NN O O
our NN O O
approach NN O I-Generic
over NN O O
state-of-the-art NN O O
change NN O I-Method
detection NN O I-Method
methods NN O I-Method
and NN O O
its NN O O
ability NN O O
to NN O O
distinguish NN O O
real NN O I-OtherScientificTerm
scene NN O I-OtherScientificTerm
changes NN O I-OtherScientificTerm
from NN O O
false NN O O
ones NN O O
caused NN O O
by NN O O
lighting NN O I-OtherScientificTerm
variations NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Automatic NN O I-Metric
evaluation NN O I-Metric
metrics NN O I-Metric
for NN O O
Machine NN O I-Task
Translation NN O I-Task
-LRB- NN O I-Task
MT NN O I-Task
-RRB- NN O I-Task
systems NN O I-Task
, NN O O
such NN O O
as NN O O
BLEU NN O I-Metric
or NN O O
NIST NN O I-Metric
, NN O O
are NN O O
now NN O O
well NN O O
established NN O O
. NN O O

Yet NN O O
, NN O O
they NN O I-Generic
are NN O O
scarcely NN O O
used NN O O
for NN O O
the NN O O
assessment NN O I-Task
of NN O I-Task
language NN O I-OtherScientificTerm
pairs NN O I-OtherScientificTerm
like NN O O
English-Chinese NN O I-OtherScientificTerm
or NN O O
English-Japanese NN O I-OtherScientificTerm
, NN O O
because NN O O
of NN O O
the NN O O
word NN O I-Task
segmentation NN O I-Task
problem NN O I-Task
. NN O O

This NN O O
study NN O O
establishes NN O O
the NN O O
equivalence NN O O
between NN O O
the NN O O
standard NN O O
use NN O O
of NN O O
BLEU NN O I-Metric
in NN O O
word NN O I-Method
n-grams NN O I-Method
and NN O O
its NN O O
application NN O O
at NN O O
the NN O O
character NN O I-OtherScientificTerm
level NN O I-OtherScientificTerm
. NN O O

The NN O O
use NN O O
of NN O O
BLEU NN O I-Metric
at NN O O
the NN O O
character NN O I-OtherScientificTerm
level NN O I-OtherScientificTerm
eliminates NN O O
the NN O O
word NN O I-Task
segmentation NN O I-Task
problem NN O I-Task
: NN O O
it NN O I-Generic
makes NN O O
it NN O O
possible NN O O
to NN O O
directly NN O O
compare NN O O
commercial NN O I-Method
systems NN O I-Method
outputting NN O O
unsegmented NN O O
texts NN O O
with NN O O
, NN O O
for NN O O
instance NN O O
, NN O O
statistical NN O I-Method
MT NN O I-Method
systems NN O I-Method
which NN O O
usually NN O O
segment NN O O
their NN O O
outputs NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
a NN O O
series NN O O
of NN O O
modifications NN O O
to NN O O
the NN O O
left NN O I-Method
corner NN O I-Method
parsing NN O I-Method
algorithm NN O I-Method
for NN O O
context-free NN O I-Method
grammars NN O I-Method
. NN O O

It NN O O
is NN O O
argued NN O O
that NN O O
the NN O O
resulting NN O O
algorithm NN O I-Generic
is NN O O
both NN O O
efficient NN O O
and NN O O
flexible NN O O
and NN O O
is NN O O
, NN O O
therefore NN O O
, NN O O
a NN O O
good NN O O
choice NN O O
for NN O O
the NN O O
parser NN O I-Method
used NN O O
in NN O O
a NN O O
natural NN O I-Task
language NN O I-Task
interface NN O I-Task
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
novel NN O O
statistical NN O I-Method
singing NN O I-Method
voice NN O I-Method
conversion NN O I-Method
-LRB- NN O I-Method
SVC NN O I-Method
-RRB- NN O I-Method
technique NN O I-Method
with NN O O
direct NN O I-Method
waveform NN O I-Method
modification NN O I-Method
based NN O O
on NN O O
the NN O O
spectrum NN O I-OtherScientificTerm
differential NN O I-OtherScientificTerm
that NN O O
can NN O O
convert NN O O
voice NN O I-OtherScientificTerm
timbre NN O I-OtherScientificTerm
of NN O O
a NN O O
source NN O O
singer NN O O
into NN O O
that NN O O
of NN O O
a NN O O
target NN O O
singer NN O O
without NN O O
using NN O O
a NN O O
vocoder NN O I-Method
to NN O O
generate NN O O
converted NN O I-OtherScientificTerm
singing NN O I-OtherScientificTerm
voice NN O I-OtherScientificTerm
waveforms NN O I-OtherScientificTerm
. NN O O

SVC NN O I-Method
makes NN O O
it NN O O
possible NN O O
to NN O O
convert NN O O
singing NN O I-OtherScientificTerm
voice NN O I-OtherScientificTerm
characteristics NN O I-OtherScientificTerm
of NN O O
an NN O O
arbitrary NN O O
source NN O O
singer NN O O
into NN O O
those NN O O
of NN O O
an NN O O
arbitrary NN O O
target NN O O
singer NN O O
. NN O O

However NN O O
, NN O O
speech NN O I-Metric
quality NN O I-Metric
of NN O O
the NN O O
converted NN O I-OtherScientificTerm
singing NN O I-OtherScientificTerm
voice NN O I-OtherScientificTerm
is NN O O
significantly NN O O
degraded NN O O
compared NN O O
to NN O O
that NN O O
of NN O O
a NN O O
natural NN O I-OtherScientificTerm
singing NN O I-OtherScientificTerm
voice NN O I-OtherScientificTerm
due NN O O
to NN O O
various NN O O
factors NN O O
, NN O O
such NN O O
as NN O O
analysis NN O O
and NN O O
modeling NN O O
errors NN O O
in NN O O
the NN O O
vocoder-based NN O I-Method
framework NN O I-Method
. NN O O

To NN O O
alleviate NN O O
this NN O O
degradation NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
statistical NN O I-Method
conversion NN O I-Method
process NN O I-Method
that NN O O
directly NN O O
modifies NN O O
the NN O O
signal NN O O
in NN O O
the NN O O
waveform NN O O
domain NN O O
by NN O O
estimating NN O O
the NN O O
difference NN O O
in NN O O
the NN O O
spectra NN O I-OtherScientificTerm
of NN O O
the NN O O
source NN O O
and NN O O
target NN O O
singers NN O O
' NN O O
singing NN O O
voices NN O O
. NN O O

The NN O O
differential NN O I-OtherScientificTerm
spectral NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
is NN O O
directly NN O O
estimated NN O O
using NN O O
a NN O O
differential NN O I-Method
Gaussian NN O I-Method
mixture NN O I-Method
model NN O I-Method
-LRB- NN O I-Method
GMM NN O I-Method
-RRB- NN O I-Method
that NN O O
is NN O O
analytically NN O O
derived NN O O
from NN O O
the NN O O
traditional NN O O
GMM NN O I-Method
used NN O O
as NN O O
a NN O O
conversion NN O I-Method
model NN O I-Method
in NN O O
the NN O O
conventional NN O O
SVC NN O I-Method
. NN O O

The NN O O
experimental NN O O
results NN O O
demonstrate NN O O
that NN O O
the NN O O
proposed NN O O
method NN O I-Generic
makes NN O O
it NN O O
possible NN O O
to NN O O
significantly NN O O
improve NN O O
speech NN O I-Metric
quality NN O I-Metric
in NN O O
the NN O O
converted NN O O
singing NN O O
voice NN O O
while NN O O
preserving NN O O
the NN O O
conversion NN O I-Metric
accuracy NN O I-Metric
of NN O I-Metric
singer NN O I-Metric
identity NN O I-Metric
compared NN O O
to NN O O
the NN O O
conventional NN O O
SVC NN O I-Method
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
state NN O O
the NN O O
challenges NN O O
of NN O O
high-level NN O I-Task
program NN O I-Task
execution NN O I-Task
in NN O O
multi-agent NN O O
settings NN O O
. NN O O

We NN O O
first NN O O
introduce NN O O
high-level NN O I-Task
program NN O I-Task
execution NN O I-Task
and NN O O
the NN O O
related NN O O
work NN O O
. NN O O

Then NN O O
we NN O O
describe NN O O
the NN O O
completed NN O O
work NN O O
, NN O O
the NN O O
future NN O O
work NN O O
and NN O O
its NN O O
approaches NN O O
. NN O O

We NN O O
conclude NN O O
with NN O O
the NN O O
expected NN O O
contributions NN O O
of NN O O
our NN O O
research NN O O
. NN O O

-DOCSTART- NN O O

During NN O O
late-2013 NN O O
through NN O O
early-2014 NN O O
NIST NN O O
coordinated NN O O
a NN O O
special NN O O
i-vector NN O I-Material
challenge NN O I-Material
based NN O O
on NN O O
data NN O O
used NN O O
in NN O O
previous NN O O
NIST NN O I-Material
Speaker NN O I-Material
Recognition NN O I-Material
Evaluations NN O I-Material
-LRB- NN O I-Material
SREs NN O I-Material
-RRB- NN O I-Material
. NN O O

Unlike NN O O
evaluations NN O O
in NN O O
the NN O O
SRE NN O I-Material
series NN O I-Material
, NN O O
the NN O O
i-vector NN O I-Material
challenge NN O I-Material
was NN O O
run NN O O
entirely NN O O
online NN O O
and NN O O
used NN O O
fixed-length NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
vectors NN O I-OtherScientificTerm
projected NN O O
into NN O O
a NN O O
low-dimensional NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
i-vectors NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
rather NN O O
than NN O O
audio NN O I-Material
recordings NN O I-Material
. NN O O

These NN O O
changes NN O O
made NN O O
the NN O O
challenge NN O O
more NN O O
readily NN O O
accessible NN O O
, NN O O
especially NN O O
to NN O O
participants NN O O
from NN O O
outside NN O O
the NN O O
audio NN O I-OtherScientificTerm
processing NN O I-OtherScientificTerm
field NN O I-OtherScientificTerm
. NN O O

Compared NN O O
to NN O O
the NN O O
2012 NN O O
SRE NN O I-Material
, NN O O
the NN O O
i-vector NN O I-Material
challenge NN O I-Material
saw NN O O
an NN O O
increase NN O O
in NN O O
the NN O O
number NN O O
of NN O O
participants NN O O
by NN O O
nearly NN O O
a NN O O
factor NN O O
of NN O O
two NN O O
, NN O O
and NN O O
a NN O O
two NN O O
orders NN O O
of NN O O
magnitude NN O O
increase NN O O
in NN O O
the NN O O
number NN O O
of NN O O
systems NN O O
submitted NN O O
for NN O O
evaluation NN O O
. NN O O

Initial NN O O
results NN O O
indicate NN O O
the NN O O
leading NN O I-Generic
system NN O I-Generic
achieved NN O O
an NN O O
approximate NN O O
37 NN O O
% NN O O
improvement NN O O
relative NN O O
to NN O O
the NN O O
baseline NN O I-Generic
system NN O I-Generic
. NN O O

-DOCSTART- NN O O

Theoretical NN O O
research NN O O
in NN O O
the NN O O
area NN O O
of NN O O
machine NN O I-Task
translation NN O I-Task
usually NN O O
involves NN O O
the NN O O
search NN O O
for NN O O
and NN O O
creation NN O O
of NN O O
an NN O O
appropriate NN O O
formalism NN O I-Generic
. NN O O

An NN O O
important NN O O
issue NN O O
in NN O O
this NN O O
respect NN O O
is NN O O
the NN O O
way NN O O
in NN O O
which NN O O
the NN O O
compositionality NN O O
of NN O O
translation NN O I-Task
is NN O O
to NN O O
be NN O O
defined NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
will NN O O
introduce NN O O
the NN O O
anaphoric NN O I-Method
component NN O I-Method
of NN O O
the NN O O
Mimo NN O I-Method
formalism NN O I-Method
. NN O O

It NN O O
makes NN O O
the NN O O
definition NN O O
and NN O O
translation NN O I-Task
of NN O I-Task
anaphoric NN O I-Task
relations NN O I-Task
possible NN O O
, NN O O
relations NN O O
which NN O O
are NN O O
usually NN O O
problematic NN O O
for NN O O
systems NN O O
that NN O O
adhere NN O O
to NN O O
strict NN O I-OtherScientificTerm
compositionality NN O I-OtherScientificTerm
. NN O O

In NN O O
Mimo NN O I-OtherScientificTerm
, NN O O
the NN O O
translation NN O I-Task
of NN O I-Task
anaphoric NN O I-Task
relations NN O I-Task
is NN O O
compositional NN O O
. NN O O

The NN O O
anaphoric NN O I-Method
component NN O I-Method
is NN O O
used NN O O
to NN O O
define NN O O
linguistic NN O I-OtherScientificTerm
phenomena NN O I-OtherScientificTerm
such NN O O
as NN O O
wh-movement NN O I-OtherScientificTerm
, NN O O
the NN O I-OtherScientificTerm
passive NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
binding NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
reflexives NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
pronouns NN O I-OtherScientificTerm
mono-lingually NN O O
. NN O O

The NN O O
actual NN O O
working NN O O
of NN O O
the NN O O
component NN O O
will NN O O
be NN O O
shown NN O O
in NN O O
this NN O O
paper NN O O
by NN O O
means NN O O
of NN O O
a NN O O
detailed NN O O
discussion NN O O
of NN O O
wh-movement NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

A NN O O
recognition NN O I-Method
scheme NN O I-Method
that NN O O
scales NN O O
efficiently NN O O
to NN O O
a NN O O
large NN O O
number NN O O
of NN O O
objects NN O O
is NN O O
presented NN O O
. NN O O

The NN O O
efficiency NN O I-Metric
and NN O O
quality NN O I-Metric
is NN O O
exhibited NN O O
in NN O O
a NN O O
live NN O O
demonstration NN O O
that NN O O
recognizes NN O O
CD-covers NN O I-Material
from NN O O
a NN O O
database NN O I-Generic
of NN O O
40000 NN O O
images NN O I-Material
of NN O I-Material
popular NN O I-Material
music NN O I-Material
CD NN O I-Material
's NN O I-Material
. NN O O

The NN O O
scheme NN O I-Generic
builds NN O O
upon NN O O
popular NN O O
techniques NN O O
of NN O O
indexing NN O I-Method
descriptors NN O I-Method
extracted NN O O
from NN O O
local NN O I-OtherScientificTerm
regions NN O I-OtherScientificTerm
, NN O O
and NN O O
is NN O O
robust NN O O
to NN O O
background NN O I-OtherScientificTerm
clutter NN O I-OtherScientificTerm
and NN O O
occlusion NN O I-OtherScientificTerm
. NN O O

The NN O O
local NN O I-Method
region NN O I-Method
descriptors NN O I-Method
are NN O O
hierarchically NN O O
quantized NN O O
in NN O O
a NN O O
vocabulary NN O I-OtherScientificTerm
tree NN O I-OtherScientificTerm
. NN O O

The NN O O
vocabulary NN O I-OtherScientificTerm
tree NN O I-OtherScientificTerm
allows NN O O
a NN O O
larger NN O O
and NN O O
more NN O O
discriminatory NN O O
vocabulary NN O O
to NN O O
be NN O O
used NN O O
efficiently NN O O
, NN O O
which NN O O
we NN O O
show NN O O
experimentally NN O O
leads NN O O
to NN O O
a NN O O
dramatic NN O O
improvement NN O O
in NN O O
retrieval NN O I-Metric
quality NN O I-Metric
. NN O O

The NN O O
most NN O O
significant NN O O
property NN O O
of NN O O
the NN O O
scheme NN O O
is NN O O
that NN O O
the NN O O
tree NN O I-OtherScientificTerm
directly NN O O
defines NN O O
the NN O O
quantization NN O I-OtherScientificTerm
. NN O O

The NN O O
quantization NN O I-OtherScientificTerm
and NN O O
the NN O O
indexing NN O I-OtherScientificTerm
are NN O O
therefore NN O O
fully NN O O
integrated NN O O
, NN O O
essentially NN O O
being NN O O
one NN O O
and NN O O
the NN O O
same NN O O
. NN O O

The NN O O
recognition NN O I-Metric
quality NN O I-Metric
is NN O O
evaluated NN O O
through NN O O
retrieval NN O I-Task
on NN O O
a NN O O
database NN O I-Material
with NN O I-Material
ground NN O I-Material
truth NN O I-Material
, NN O O
showing NN O O
the NN O O
power NN O O
of NN O O
the NN O O
vocabulary NN O I-Method
tree NN O I-Method
approach NN O I-Method
, NN O O
going NN O O
as NN O O
high NN O O
as NN O O
1 NN O O
million NN O O
images NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
method NN O I-Generic
for NN O O
blind NN O I-Task
estimation NN O I-Task
of NN O I-Task
reverberation NN O I-Task
times NN O I-Task
in NN O O
reverberant NN O I-OtherScientificTerm
enclosures NN O I-OtherScientificTerm
. NN O O

The NN O O
proposed NN O O
algorithm NN O I-Generic
is NN O O
based NN O O
on NN O O
a NN O O
statistical NN O I-Method
model NN O I-Method
of NN O I-Method
short-term NN O I-Method
log-energy NN O I-Method
sequences NN O I-Method
for NN O O
echo-free NN O I-Material
speech NN O I-Material
. NN O O

Given NN O O
a NN O O
speech NN O O
utterance NN O O
recorded NN O O
in NN O O
a NN O O
reverberant NN O O
room NN O O
, NN O O
it NN O O
computes NN O O
a NN O O
Maximum NN O I-Method
Likelihood NN O I-Method
estimate NN O I-Method
of NN O I-Method
the NN O I-Method
room NN O I-Method
full-band NN O I-Method
reverberation NN O I-Method
time NN O I-Method
. NN O O

The NN O O
estimation NN O I-Generic
method NN O I-Generic
is NN O O
shown NN O O
to NN O O
require NN O O
little NN O O
data NN O O
and NN O O
to NN O O
perform NN O O
satisfactorily NN O O
. NN O O

The NN O O
method NN O I-Generic
has NN O O
been NN O O
successfully NN O O
applied NN O O
to NN O O
robust NN O I-Task
automatic NN O I-Task
speech NN O I-Task
recognition NN O I-Task
in NN O O
reverberant NN O I-OtherScientificTerm
environments NN O I-OtherScientificTerm
by NN O O
model NN O I-Method
selection NN O I-Method
. NN O O

For NN O O
this NN O O
application NN O I-Generic
, NN O O
the NN O O
reverberation NN O I-OtherScientificTerm
time NN O I-OtherScientificTerm
is NN O O
first NN O O
estimated NN O O
from NN O O
the NN O O
reverberated NN O I-OtherScientificTerm
speech NN O I-OtherScientificTerm
utterance NN O I-OtherScientificTerm
to NN O O
be NN O O
recognized NN O O
. NN O O

The NN O O
estimation NN O I-Generic
is NN O O
then NN O O
used NN O O
to NN O O
select NN O O
the NN O O
best NN O O
acoustic NN O I-Method
model NN O I-Method
out NN O O
of NN O O
a NN O O
library NN O O
of NN O O
models NN O I-Generic
trained NN O O
in NN O O
various NN O O
artificial NN O I-OtherScientificTerm
re-verberant NN O I-OtherScientificTerm
conditions NN O I-OtherScientificTerm
. NN O O

Speech NN O I-Task
recognition NN O I-Task
experiments NN O O
in NN O O
simulated NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
real NN O I-OtherScientificTerm
reverberant NN O I-OtherScientificTerm
environments NN O I-OtherScientificTerm
show NN O O
the NN O O
efficiency NN O O
of NN O O
our NN O O
approach NN O I-Generic
which NN O O
outperforms NN O O
standard NN O O
channel NN O I-Method
normaliza-tion NN O I-Method
techniques NN O I-Method
. NN O O

-DOCSTART- NN O O

Computer NN O I-Generic
programs NN O I-Generic
so NN O O
far NN O O
have NN O O
not NN O O
fared NN O O
well NN O O
in NN O O
modeling NN O O
language NN O I-Task
acquisition NN O I-Task
. NN O O

For NN O O
one NN O O
thing NN O O
, NN O O
learning NN O I-Method
methodology NN O I-Method
applicable NN O O
in NN O O
general NN O I-Material
domains NN O I-Material
does NN O O
not NN O O
readily NN O O
lend NN O O
itself NN O O
in NN O O
the NN O O
linguistic NN O I-Material
domain NN O I-Material
. NN O O

For NN O O
another NN O O
, NN O O
linguistic NN O I-Method
representation NN O I-Method
used NN O O
by NN O O
language NN O I-Method
processing NN O I-Method
systems NN O I-Method
is NN O O
not NN O O
geared NN O O
to NN O O
learning NN O O
. NN O O

We NN O O
introduced NN O O
a NN O O
new NN O O
linguistic NN O I-Method
representation NN O I-Method
, NN O O
the NN O O
Dynamic NN O I-OtherScientificTerm
Hierarchical NN O I-OtherScientificTerm
Phrasal NN O I-OtherScientificTerm
Lexicon NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
DHPL NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
-LSB- NN O O
Zernik88 NN O O
-RSB- NN O O
, NN O O
to NN O O
facilitate NN O O
language NN O I-Task
acquisition NN O I-Task
. NN O O

From NN O O
this NN O O
, NN O O
a NN O O
language NN O I-Method
learning NN O I-Method
model NN O I-Method
was NN O O
implemented NN O O
in NN O O
the NN O O
program NN O O
RINA NN O I-Method
, NN O O
which NN O O
enhances NN O O
its NN O O
own NN O O
lexical NN O I-OtherScientificTerm
hierarchy NN O I-OtherScientificTerm
by NN O O
processing NN O O
examples NN O O
in NN O O
context NN O O
. NN O O

We NN O O
identified NN O O
two NN O O
tasks NN O O
: NN O O
First NN O O
, NN O O
how NN O O
linguistic NN O I-OtherScientificTerm
concepts NN O I-OtherScientificTerm
are NN O O
acquired NN O O
from NN O O
training NN O O
examples NN O O
and NN O O
organized NN O O
in NN O O
a NN O O
hierarchy NN O I-Generic
; NN O O
this NN O O
task NN O O
was NN O O
discussed NN O O
in NN O O
previous NN O O
papers NN O O
-LSB- NN O O
Zernik87 NN O O
-RSB- NN O O
. NN O O

Second NN O O
, NN O O
we NN O O
show NN O O
in NN O O
this NN O O
paper NN O O
how NN O O
a NN O O
lexical NN O I-OtherScientificTerm
hierarchy NN O I-OtherScientificTerm
is NN O O
used NN O O
in NN O O
predicting NN O O
new NN O O
linguistic NN O I-OtherScientificTerm
concepts NN O I-OtherScientificTerm
. NN O O

Thus NN O O
, NN O O
a NN O O
program NN O I-Generic
does NN O O
not NN O O
stall NN O O
even NN O O
in NN O O
the NN O O
presence NN O O
of NN O O
a NN O O
lexical NN O O
unknown NN O O
, NN O O
and NN O O
a NN O O
hypothesis NN O O
can NN O O
be NN O O
produced NN O O
for NN O O
covering NN O O
that NN O O
lexical NN O O
gap NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
novel NN O O
ensemble NN O I-Method
learning NN O I-Method
approach NN O I-Method
to NN O O
resolving NN O O
German NN O I-OtherScientificTerm
pronouns NN O I-OtherScientificTerm
. NN O O

Boosting NN O I-Method
, NN O O
the NN O O
method NN O O
in NN O O
question NN O O
, NN O O
combines NN O O
the NN O O
moderately NN O O
accurate NN O O
hypotheses NN O O
of NN O O
several NN O O
classifiers NN O I-Method
to NN O O
form NN O O
a NN O O
highly NN O O
accurate NN O O
one NN O O
. NN O O

Experiments NN O O
show NN O O
that NN O O
this NN O O
approach NN O I-Generic
is NN O O
superior NN O O
to NN O O
a NN O O
single NN O O
decision-tree NN O I-Method
classifier NN O I-Method
. NN O O

Furthermore NN O O
, NN O O
we NN O O
present NN O O
a NN O O
standalone NN O I-Method
system NN O I-Method
that NN O O
resolves NN O O
pronouns NN O I-OtherScientificTerm
in NN O O
unannotated NN O I-Material
text NN O I-Material
by NN O O
using NN O O
a NN O O
fully NN O O
automatic NN O O
sequence NN O O
of NN O O
preprocessing NN O I-Method
modules NN O I-Method
that NN O O
mimics NN O O
the NN O O
manual NN O I-Task
annotation NN O I-Task
process NN O I-Task
. NN O O

Although NN O O
the NN O O
system NN O I-Generic
performs NN O O
well NN O O
within NN O O
a NN O O
limited NN O O
textual NN O I-Material
domain NN O I-Material
, NN O O
further NN O O
research NN O O
is NN O O
needed NN O O
to NN O O
make NN O O
it NN O I-Generic
effective NN O O
for NN O O
open-domain NN O I-Task
question NN O I-Task
answering NN O I-Task
and NN O O
text NN O I-Task
summarisation NN O I-Task
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
compare NN O O
the NN O O
performance NN O O
of NN O O
a NN O O
state-of-the-art NN O O
statistical NN O I-Method
parser NN O I-Method
-LRB- NN O O
Bikel NN O O
, NN O O
2004 NN O O
-RRB- NN O O
in NN O O
parsing NN O I-Task
written NN O I-Material
and NN O I-Material
spoken NN O I-Material
language NN O I-Material
and NN O O
in NN O O
generating NN O I-Task
sub-categorization NN O I-Task
cues NN O I-Task
from NN O O
written NN O I-Material
and NN O I-Material
spoken NN O I-Material
language NN O I-Material
. NN O O

Although NN O O
Bikel NN O I-Method
's NN O I-Method
parser NN O I-Method
achieves NN O O
a NN O O
higher NN O O
accuracy NN O I-Metric
for NN O O
parsing NN O I-Task
written NN O I-Material
language NN O I-Material
, NN O O
it NN O I-Generic
achieves NN O O
a NN O O
higher NN O O
accuracy NN O I-Metric
when NN O O
extracting NN O O
subcategorization NN O I-OtherScientificTerm
cues NN O I-OtherScientificTerm
from NN O O
spoken NN O I-Material
language NN O I-Material
. NN O O

Our NN O O
experiments NN O O
also NN O O
show NN O O
that NN O O
current NN O O
technology NN O I-Generic
for NN O O
extracting NN O I-Task
subcategorization NN O I-Task
frames NN O I-Task
initially NN O O
designed NN O O
for NN O O
written NN O I-Material
texts NN O I-Material
works NN O O
equally NN O O
well NN O O
for NN O O
spoken NN O I-Material
language NN O I-Material
. NN O O

Additionally NN O O
, NN O O
we NN O O
explore NN O O
the NN O O
utility NN O O
of NN O O
punctuation NN O I-OtherScientificTerm
in NN O O
helping NN O O
parsing NN O I-Method
and NN O O
extraction NN O I-Task
of NN O I-Task
subcategorization NN O I-Task
cues NN O I-Task
. NN O O

Our NN O O
experiments NN O O
show NN O O
that NN O O
punctuation NN O I-OtherScientificTerm
is NN O O
of NN O O
little NN O O
help NN O O
in NN O O
parsing NN O I-Task
spoken NN O I-Material
language NN O I-Material
and NN O O
extracting NN O I-Task
subcategorization NN O I-OtherScientificTerm
cues NN O I-OtherScientificTerm
from NN O O
spoken NN O I-Material
language NN O I-Material
. NN O O

This NN O O
indicates NN O O
that NN O O
there NN O O
is NN O O
no NN O O
need NN O O
to NN O O
add NN O O
punctuation NN O I-OtherScientificTerm
in NN O O
transcribing NN O O
spoken NN O I-Material
corpora NN O I-Material
simply NN O O
in NN O O
order NN O O
to NN O O
help NN O O
parsers NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
an NN O O
alignment NN O I-Method
adaptation NN O I-Method
approach NN O I-Method
to NN O O
improve NN O O
domain-specific NN O I-Task
-LRB- NN O I-Task
in-domain NN O I-Task
-RRB- NN O I-Task
word NN O I-Task
alignment NN O I-Task
. NN O O

The NN O O
basic NN O O
idea NN O O
of NN O O
alignment NN O I-Method
adaptation NN O I-Method
is NN O O
to NN O O
use NN O O
out-of-domain NN O I-Material
corpus NN O I-Material
to NN O O
improve NN O O
in-domain NN O I-Task
word NN O I-Task
alignment NN O I-Task
results NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
first NN O O
train NN O O
two NN O O
statistical NN O I-Method
word NN O I-Method
alignment NN O I-Method
models NN O I-Method
with NN O O
the NN O O
large-scale NN O I-Material
out-of-domain NN O I-Material
corpus NN O I-Material
and NN O O
the NN O O
small-scale NN O I-Material
in-domain NN O I-Material
corpus NN O I-Material
respectively NN O O
, NN O O
and NN O O
then NN O O
interpolate NN O O
these NN O O
two NN O O
models NN O I-Generic
to NN O O
improve NN O O
the NN O O
domain-specific NN O I-Task
word NN O I-Task
alignment NN O I-Task
. NN O O

Experimental NN O O
results NN O O
show NN O O
that NN O O
our NN O O
approach NN O I-Generic
improves NN O O
domain-specific NN O I-Task
word NN O I-Task
alignment NN O I-Task
in NN O O
terms NN O O
of NN O O
both NN O O
precision NN O I-Metric
and NN O O
recall NN O I-Metric
, NN O O
achieving NN O O
a NN O O
relative NN O I-Metric
error NN O I-Metric
rate NN O I-Metric
reduction NN O I-Metric
of NN O O
6.56 NN O O
% NN O O
as NN O O
compared NN O O
with NN O O
the NN O O
state-of-the-art NN O I-Generic
technologies NN O I-Generic
. NN O O

-DOCSTART- NN O O

With NN O O
performance NN O O
above NN O O
97 NN O O
% NN O O
accuracy NN O I-Metric
for NN O O
newspaper NN O I-Material
text NN O I-Material
, NN O O
part NN O I-Task
of NN O I-Task
speech NN O I-Task
-LRB- NN O I-Task
pos NN O I-Task
-RRB- NN O I-Task
tagging NN O I-Task
might NN O O
be NN O O
considered NN O O
a NN O O
solved NN O O
problem NN O O
. NN O O

Previous NN O O
studies NN O O
have NN O O
shown NN O O
that NN O O
allowing NN O O
the NN O O
parser NN O I-Method
to NN O O
resolve NN O O
pos NN O I-OtherScientificTerm
tag NN O I-OtherScientificTerm
ambiguity NN O I-OtherScientificTerm
does NN O O
not NN O O
improve NN O O
performance NN O O
. NN O O

However NN O O
, NN O O
for NN O O
grammar NN O I-Method
formalisms NN O I-Method
which NN O O
use NN O O
more NN O O
fine-grained NN O I-OtherScientificTerm
grammatical NN O I-OtherScientificTerm
categories NN O I-OtherScientificTerm
, NN O O
for NN O O
example NN O O
tag NN O I-OtherScientificTerm
and NN O O
ccg NN O I-OtherScientificTerm
, NN O O
tagging NN O I-Metric
accuracy NN O I-Metric
is NN O O
much NN O O
lower NN O O
. NN O O

In NN O O
fact NN O O
, NN O O
for NN O O
these NN O O
formalisms NN O I-Generic
, NN O O
premature NN O I-OtherScientificTerm
ambiguity NN O I-OtherScientificTerm
resolution NN O I-OtherScientificTerm
makes NN O O
parsing NN O I-Task
infeasible NN O O
. NN O O

We NN O O
describe NN O O
a NN O O
multi-tagging NN O I-Method
approach NN O I-Method
which NN O O
maintains NN O O
a NN O O
suitable NN O O
level NN O O
of NN O O
lexical NN O I-OtherScientificTerm
category NN O I-OtherScientificTerm
ambiguity NN O I-OtherScientificTerm
for NN O O
accurate NN O O
and NN O O
efficient NN O O
ccg NN O I-Task
parsing NN O I-Task
. NN O O

We NN O O
extend NN O O
this NN O O
multi-tagging NN O I-Method
approach NN O I-Method
to NN O O
the NN O O
pos NN O I-OtherScientificTerm
level NN O I-OtherScientificTerm
to NN O O
overcome NN O O
errors NN O O
introduced NN O O
by NN O O
automatically NN O O
assigned NN O O
pos NN O I-OtherScientificTerm
tags NN O I-OtherScientificTerm
. NN O O

Although NN O O
pos NN O I-Metric
tagging NN O I-Metric
accuracy NN O I-Metric
seems NN O O
high NN O O
, NN O O
maintaining NN O O
some NN O O
pos NN O I-OtherScientificTerm
tag NN O I-OtherScientificTerm
ambiguity NN O I-OtherScientificTerm
in NN O O
the NN O O
language NN O I-Method
processing NN O I-Method
pipeline NN O I-Method
results NN O O
in NN O O
more NN O O
accurate NN O O
ccg NN O I-Method
supertagging NN O I-Method
. NN O O

-DOCSTART- NN O O

We NN O O
previously NN O O
presented NN O O
a NN O O
framework NN O I-Generic
for NN O O
segmentation NN O I-Task
of NN O I-Task
complex NN O I-Task
scenes NN O I-Task
using NN O O
multiple NN O O
physical NN O I-OtherScientificTerm
hypotheses NN O I-OtherScientificTerm
for NN O O
simple NN O I-OtherScientificTerm
image NN O I-OtherScientificTerm
regions NN O I-OtherScientificTerm
. NN O O

A NN O O
consequence NN O O
of NN O O
that NN O O
framework NN O I-Generic
was NN O O
a NN O O
proposal NN O O
for NN O O
a NN O O
new NN O O
approach NN O I-Generic
to NN O O
the NN O O
segmentation NN O I-Task
of NN O I-Task
complex NN O I-Task
scenes NN O I-Task
into NN O O
regions NN O O
corresponding NN O O
to NN O O
coherent NN O I-OtherScientificTerm
surfaces NN O I-OtherScientificTerm
rather NN O O
than NN O O
merely NN O O
regions NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
similar NN O I-OtherScientificTerm
color NN O I-OtherScientificTerm
. NN O O

Herein NN O O
we NN O O
present NN O O
an NN O O
implementation NN O O
of NN O O
this NN O O
new NN O O
approach NN O I-Generic
and NN O O
show NN O O
example NN O O
segmentations NN O I-OtherScientificTerm
for NN O O
scenes NN O I-Material
containing NN O O
multi-colored NN O I-Material
piece-wise NN O I-Material
uniform NN O I-Material
objects NN O I-Material
. NN O O

Using NN O O
our NN O O
approach NN O I-Generic
we NN O O
are NN O O
able NN O O
to NN O O
intelligently NN O O
segment NN O O
scenes NN O O
with NN O O
objects NN O O
of NN O O
greater NN O O
complexity NN O I-Metric
than NN O O
previous NN O O
physics-based NN O I-Method
segmentation NN O I-Method
algorithms NN O I-Method
. NN O O

The NN O O
results NN O O
show NN O O
that NN O O
by NN O O
using NN O O
general NN O O
physical NN O I-Method
models NN O I-Method
we NN O O
obtain NN O O
segmentations NN O O
that NN O O
correspond NN O O
more NN O O
closely NN O O
to NN O O
coherent NN O I-OtherScientificTerm
surfaces NN O I-OtherScientificTerm
in NN O O
the NN O O
scene NN O O
than NN O O
segmentations NN O O
found NN O O
using NN O O
only NN O O
color NN O O
. NN O O

-DOCSTART- NN O O

SmartKom NN O I-Method
is NN O O
a NN O O
multimodal NN O I-Method
dialog NN O I-Method
system NN O I-Method
that NN O O
combines NN O O
speech NN O I-Material
, NN O O
gesture NN O I-Material
, NN O O
and NN O O
mimics NN O O
input NN O O
and NN O O
output NN O O
. NN O O

Spontaneous NN O I-Task
speech NN O I-Task
understanding NN O I-Task
is NN O O
combined NN O O
with NN O O
the NN O O
video-based NN O I-Task
recognition NN O I-Task
of NN O I-Task
natural NN O I-Task
gestures NN O I-Task
. NN O O

One NN O O
of NN O O
the NN O O
major NN O O
scientific NN O O
goals NN O O
of NN O O
SmartKom NN O I-Method
is NN O O
to NN O O
design NN O O
new NN O O
computational NN O I-Method
methods NN O I-Method
for NN O O
the NN O O
seamless NN O O
integration NN O I-Task
and NN O I-Task
mutual NN O I-Task
disambiguation NN O I-Task
of NN O I-Task
multimodal NN O I-Task
input NN O I-Task
and NN O I-Task
output NN O I-Task
on NN O O
a NN O O
semantic NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
pragmatic NN O I-OtherScientificTerm
level NN O I-OtherScientificTerm
. NN O O

SmartKom NN O I-OtherScientificTerm
is NN O O
based NN O O
on NN O O
the NN O O
situated NN O I-Method
delegation-oriented NN O I-Method
dialog NN O I-Method
paradigm NN O I-Method
, NN O O
in NN O O
which NN O O
the NN O O
user NN O O
delegates NN O O
a NN O O
task NN O O
to NN O O
a NN O O
virtual NN O I-OtherScientificTerm
communication NN O I-OtherScientificTerm
assistant NN O I-OtherScientificTerm
, NN O O
visualized NN O O
as NN O O
a NN O O
lifelike NN O O
character NN O O
on NN O O
a NN O O
graphical NN O I-OtherScientificTerm
display NN O I-OtherScientificTerm
. NN O O

We NN O O
describe NN O O
the NN O O
SmartKom NN O I-Method
architecture NN O I-Method
, NN O O
the NN O O
use NN O O
of NN O O
an NN O O
XML-based NN O I-OtherScientificTerm
markup NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
for NN O O
multimodal NN O I-Material
content NN O I-Material
, NN O O
and NN O O
some NN O O
of NN O O
the NN O O
distinguishing NN O O
features NN O O
of NN O O
the NN O O
first NN O O
fully NN O O
operational NN O O
SmartKom NN O I-Method
demonstrator NN O I-Method
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
a NN O O
single-image NN O I-Method
highlight NN O I-Method
removal NN O I-Method
method NN O I-Method
that NN O O
incorporates NN O O
illumination-based NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
into NN O O
image NN O I-Task
in-painting NN O I-Task
. NN O O

Unlike NN O O
occluded NN O I-OtherScientificTerm
image NN O I-OtherScientificTerm
regions NN O I-OtherScientificTerm
filled NN O O
by NN O O
traditional NN O O
inpainting NN O I-OtherScientificTerm
, NN O O
highlight NN O I-OtherScientificTerm
pixels NN O I-OtherScientificTerm
contain NN O O
some NN O O
useful NN O O
information NN O O
for NN O O
guiding NN O O
the NN O O
inpainting NN O I-Task
process NN O I-Task
. NN O O

Constraints NN O I-OtherScientificTerm
provided NN O O
by NN O O
observed NN O O
pixel NN O I-OtherScientificTerm
colors NN O I-OtherScientificTerm
, NN O O
highlight NN O I-OtherScientificTerm
color NN O I-OtherScientificTerm
analysis NN O I-OtherScientificTerm
and NN O O
illumination NN O I-OtherScientificTerm
color NN O I-OtherScientificTerm
uniformity NN O I-OtherScientificTerm
are NN O O
employed NN O O
in NN O O
our NN O O
method NN O I-Generic
to NN O O
improve NN O O
estimation NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
underlying NN O I-OtherScientificTerm
diffuse NN O I-OtherScientificTerm
color NN O I-OtherScientificTerm
. NN O O

The NN O O
inclusion NN O O
of NN O O
these NN O O
illumination NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
allows NN O O
for NN O O
better NN O O
recovery NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
shading NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
textures NN O I-OtherScientificTerm
by NN O O
inpainting NN O I-Task
. NN O O

Experimental NN O O
results NN O O
are NN O O
given NN O O
to NN O O
demonstrate NN O O
the NN O O
performance NN O O
of NN O O
our NN O O
method NN O I-Method
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
novel NN O O
method NN O I-Generic
, NN O O
called NN O O
local NN O I-Method
non-negative NN O I-Method
matrix NN O I-Method
factorization NN O I-Method
-LRB- NN O I-Method
LNMF NN O I-Method
-RRB- NN O I-Method
, NN O O
for NN O O
learning NN O O
spatially NN O I-Task
localized NN O I-Task
, NN O I-Task
parts-based NN O I-Task
subspace NN O I-Task
representation NN O I-Task
of NN O I-Task
visual NN O I-Task
patterns NN O I-Task
. NN O O

An NN O O
objective NN O I-OtherScientificTerm
function NN O I-OtherScientificTerm
is NN O O
defined NN O O
to NN O O
impose NN O O
lo-calization NN O I-OtherScientificTerm
constraint NN O I-OtherScientificTerm
, NN O O
in NN O O
addition NN O O
to NN O O
the NN O O
non-negativity NN O I-OtherScientificTerm
constraint NN O I-OtherScientificTerm
in NN O O
the NN O O
standard NN O O
NMF NN O I-Method
-LSB- NN O O
1 NN O O
-RSB- NN O O
. NN O O

This NN O O
gives NN O O
a NN O O
set NN O O
of NN O O
bases NN O O
which NN O O
not NN O O
only NN O O
allows NN O O
a NN O O
non-subtractive NN O I-Method
-LRB- NN O I-Method
part-based NN O I-Method
-RRB- NN O I-Method
representation NN O I-Method
of NN O I-Method
images NN O I-Method
but NN O O
also NN O O
manifests NN O O
localized NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
. NN O O

An NN O O
algorithm NN O I-Generic
is NN O O
presented NN O O
for NN O O
the NN O O
learning NN O I-Task
of NN O O
such NN O O
basis NN O O
components NN O O
. NN O O

Experimental NN O O
results NN O O
are NN O O
presented NN O O
to NN O O
compare NN O O
LNMF NN O I-Method
with NN O O
the NN O O
NMF NN O I-Method
and NN O I-Method
PCA NN O I-Method
methods NN O I-Method
for NN O O
face NN O I-Task
representation NN O I-Task
and NN O I-Task
recognition NN O I-Task
, NN O O
which NN O O
demonstrates NN O O
advantages NN O O
of NN O O
LNMF NN O I-Method
. NN O O

-DOCSTART- NN O O

Helping NN O O
end NN O O
users NN O O
build NN O O
and NN O O
check NN O O
process NN O I-Method
models NN O I-Method
is NN O O
a NN O O
challenge NN O O
for NN O O
many NN O O
science NN O I-Task
and NN O I-Task
engineering NN O I-Task
fields NN O I-Task
. NN O O

Many NN O O
AI NN O O
researchers NN O O
have NN O O
investigated NN O O
useful NN O O
ways NN O O
of NN O O
verifying NN O O
and NN O O
validating NN O O
knowledge NN O I-Material
bases NN O I-Material
for NN O O
ontologies NN O I-OtherScientificTerm
and NN O O
rules NN O I-OtherScientificTerm
, NN O O
but NN O O
it NN O O
is NN O O
not NN O O
easy NN O O
to NN O O
directly NN O O
apply NN O O
them NN O O
to NN O O
checking NN O I-Task
process NN O I-Method
models NN O I-Method
. NN O O

Other NN O O
techniques NN O O
developed NN O O
for NN O O
checking NN O I-Task
and NN O I-Task
refining NN O I-Task
planning NN O I-Task
knowledge NN O I-Task
tend NN O O
to NN O O
focus NN O O
on NN O O
automated NN O I-Task
plan NN O I-Task
generation NN O I-Task
rather NN O O
than NN O O
helping NN O O
users NN O O
author NN O O
process NN O O
information NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
complementary NN O I-Generic
approach NN O I-Generic
which NN O O
helps NN O O
users NN O O
author NN O O
and NN O O
check NN O O
process NN O I-Method
models NN O I-Method
. NN O O

Our NN O O
system NN O I-Generic
, NN O O
called NN O O
KANAL NN O I-Method
, NN O O
relates NN O O
pieces NN O O
of NN O O
information NN O O
in NN O O
process NN O I-Method
models NN O I-Method
among NN O O
themselves NN O O
and NN O O
to NN O O
the NN O O
existing NN O O
KB NN O I-Material
, NN O O
analyzing NN O O
how NN O O
different NN O O
pieces NN O O
of NN O O
input NN O O
are NN O O
put NN O O
together NN O O
to NN O O
achieve NN O O
some NN O O
effect NN O O
. NN O O

It NN O I-Generic
builds NN O O
interdepen-dency NN O I-Method
models NN O I-Method
from NN O O
this NN O O
analysis NN O O
and NN O O
uses NN O O
them NN O I-Generic
to NN O O
find NN O O
errors NN O I-OtherScientificTerm
and NN O O
propose NN O O
fixes NN O I-OtherScientificTerm
. NN O O

Our NN O O
initial NN O O
evaluation NN O O
shows NN O O
that NN O O
KANAL NN O I-Method
was NN O O
able NN O O
to NN O O
find NN O O
most NN O O
of NN O O
the NN O O
errors NN O O
in NN O O
the NN O O
process NN O I-Method
models NN O I-Method
and NN O O
suggest NN O O
useful NN O O
fixes NN O O
including NN O O
the NN O O
fixes NN O O
that NN O O
directly NN O O
point NN O O
to NN O O
the NN O O
sources NN O O
of NN O O
the NN O O
errors NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
describe NN O O
the NN O O
research NN O O
using NN O O
machine NN O I-Method
learning NN O I-Method
techniques NN O I-Method
to NN O O
build NN O O
a NN O O
comma NN O I-Method
checker NN O I-Method
to NN O O
be NN O O
integrated NN O O
in NN O O
a NN O O
grammar NN O I-Method
checker NN O I-Method
for NN O O
Basque NN O I-Material
. NN O O

After NN O O
several NN O O
experiments NN O O
, NN O O
and NN O O
trained NN O O
with NN O O
a NN O O
little NN O O
corpus NN O O
of NN O O
100,000 NN O O
words NN O O
, NN O O
the NN O O
system NN O I-Generic
guesses NN O O
correctly NN O O
not NN O O
placing NN O O
commas NN O O
with NN O O
a NN O O
precision NN O I-Metric
of NN O O
96 NN O O
% NN O O
and NN O O
a NN O O
recall NN O I-Metric
of NN O O
98 NN O O
% NN O O
. NN O O

It NN O I-Generic
also NN O O
gets NN O O
a NN O O
precision NN O I-Metric
of NN O O
70 NN O O
% NN O O
and NN O O
a NN O O
recall NN O I-Metric
of NN O O
49 NN O O
% NN O O
in NN O O
the NN O O
task NN O O
of NN O O
placing NN O I-Task
commas NN O I-Task
. NN O O

Finally NN O O
, NN O O
we NN O O
have NN O O
shown NN O O
that NN O O
these NN O O
results NN O O
can NN O O
be NN O O
improved NN O O
using NN O O
a NN O O
bigger NN O O
and NN O O
a NN O O
more NN O O
homogeneous NN O O
corpus NN O O
to NN O O
train NN O O
, NN O O
that NN O O
is NN O O
, NN O O
a NN O O
bigger NN O O
corpus NN O O
written NN O O
by NN O O
one NN O O
unique NN O O
author NN O O
. NN O O

-DOCSTART- NN O O

The NN O O
present NN O O
paper NN O O
reports NN O O
on NN O O
a NN O O
preparatory NN O O
research NN O O
for NN O O
building NN O O
a NN O O
language NN O I-Material
corpus NN O I-Material
annotation NN O I-Material
scenario NN O I-Material
capturing NN O O
the NN O O
discourse NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
in NN O O
Czech NN O I-Material
. NN O O

We NN O O
primarily NN O O
focus NN O O
on NN O O
the NN O O
description NN O O
of NN O O
the NN O O
syntactically NN O I-OtherScientificTerm
motivated NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
in NN O I-OtherScientificTerm
discourse NN O I-OtherScientificTerm
, NN O O
basing NN O O
our NN O O
findings NN O O
on NN O O
the NN O O
theoretical NN O O
background NN O O
of NN O O
the NN O O
Prague NN O I-Material
Dependency NN O I-Material
Treebank NN O I-Material
2.0 NN O I-Material
and NN O O
the NN O O
Penn NN O I-Material
Discourse NN O I-Material
Treebank NN O I-Material
2 NN O I-Material
. NN O O

Our NN O O
aim NN O O
is NN O O
to NN O O
revisit NN O O
the NN O O
present-day NN O O
syntactico-semantic NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
tectogrammatical NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
annotation NN O I-OtherScientificTerm
in NN O O
the NN O O
Prague NN O I-Material
Dependency NN O I-Material
Treebank NN O I-Material
, NN O O
extend NN O O
it NN O I-Generic
for NN O O
the NN O O
purposes NN O O
of NN O O
a NN O O
sentence-boundary-crossing NN O I-Task
representation NN O I-Task
and NN O O
eventually NN O O
to NN O O
design NN O O
a NN O O
new NN O O
, NN O O
discourse NN O I-Task
level NN O I-Task
of NN O I-Task
annotation NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
feasible NN O O
process NN O O
of NN O O
such NN O O
a NN O O
transfer NN O O
, NN O O
comparing NN O O
the NN O O
possibilities NN O O
the NN O O
Praguian NN O I-Method
dependency-based NN O I-Method
approach NN O I-Method
offers NN O O
with NN O O
the NN O O
Penn NN O I-Method
discourse NN O I-Method
annotation NN O I-Method
based NN O O
primarily NN O O
on NN O O
the NN O O
analysis NN O I-Method
and NN O I-Method
classification NN O I-Method
of NN O I-Method
discourse NN O I-Method
connectives NN O I-Method
. NN O O

-DOCSTART- NN O O

Regression-based NN O I-Method
techniques NN O I-Method
have NN O O
shown NN O O
promising NN O O
results NN O O
for NN O O
people NN O I-Task
counting NN O I-Task
in NN O I-Task
crowded NN O I-Task
scenes NN O I-Task
. NN O O

However NN O O
, NN O O
most NN O O
existing NN O O
techniques NN O I-Generic
require NN O O
expensive NN O O
and NN O O
laborious NN O O
data NN O I-Task
annotation NN O I-Task
for NN O O
model NN O I-Task
training NN O I-Task
. NN O O

In NN O O
this NN O O
study NN O O
, NN O O
we NN O O
propose NN O O
to NN O O
address NN O O
this NN O O
problem NN O O
from NN O O
three NN O O
perspectives NN O O
: NN O O
-LRB- NN O O
1 NN O O
-RRB- NN O O
Instead NN O O
of NN O O
exhaustively NN O O
annotating NN O O
every NN O O
single NN O O
frame NN O O
, NN O O
the NN O O
most NN O O
informative NN O I-OtherScientificTerm
frames NN O I-OtherScientificTerm
are NN O O
selected NN O O
for NN O O
annotation NN O I-Task
automatically NN O O
and NN O O
actively NN O O
. NN O O

-LRB- NN O O
2 NN O O
-RRB- NN O O
Rather NN O O
than NN O O
learning NN O O
from NN O O
only NN O O
labelled NN O I-Generic
data NN O I-Generic
, NN O O
the NN O O
abundant NN O I-Material
unlabelled NN O I-Material
data NN O I-Material
are NN O O
exploited NN O O
. NN O O

-LRB- NN O O
3 NN O O
-RRB- NN O O
Labelled NN O O
data NN O O
from NN O O
other NN O O
scenes NN O O
are NN O O
employed NN O O
to NN O O
further NN O O
alleviate NN O O
the NN O O
burden NN O O
for NN O O
data NN O I-Task
annotation NN O I-Task
. NN O O

All NN O O
three NN O O
ideas NN O O
are NN O O
implemented NN O O
in NN O O
a NN O O
unified NN O I-Method
active NN O I-Method
and NN O I-Method
semi-supervised NN O I-Method
regression NN O I-Method
framework NN O I-Method
with NN O O
ability NN O O
to NN O O
perform NN O O
transfer NN O I-Method
learning NN O I-Method
, NN O O
by NN O O
exploiting NN O O
the NN O O
underlying NN O O
geometric NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
crowd NN O I-OtherScientificTerm
patterns NN O I-OtherScientificTerm
via NN O O
manifold NN O I-Method
analysis NN O I-Method
. NN O O

Extensive NN O O
experiments NN O O
validate NN O O
the NN O O
effectiveness NN O O
of NN O O
our NN O O
approach NN O I-Generic
. NN O O

-DOCSTART- NN O O

Representing NN O I-Method
images NN O I-Method
with NN O I-Method
layers NN O I-Method
has NN O O
many NN O O
important NN O O
applications NN O I-Generic
, NN O O
such NN O O
as NN O O
video NN O I-Task
compression NN O I-Task
, NN O O
motion NN O I-Task
analysis NN O I-Task
, NN O O
and NN O O
3D NN O I-Task
scene NN O I-Task
analysis NN O I-Task
. NN O O

This NN O O
paper NN O O
presents NN O O
an NN O O
approach NN O I-Generic
to NN O O
reliably NN O O
extracting NN O O
layers NN O I-OtherScientificTerm
from NN O O
images NN O I-Material
by NN O O
taking NN O O
advantages NN O O
of NN O O
the NN O O
fact NN O O
that NN O O
homographies NN O I-OtherScientificTerm
induced NN O O
by NN O O
planar NN O I-OtherScientificTerm
patches NN O I-OtherScientificTerm
in NN O O
the NN O O
scene NN O I-OtherScientificTerm
form NN O O
a NN O O
low NN O I-OtherScientificTerm
dimensional NN O I-OtherScientificTerm
linear NN O I-OtherScientificTerm
subspace NN O I-OtherScientificTerm
. NN O O

Layers NN O I-OtherScientificTerm
in NN O O
the NN O O
input NN O O
images NN O I-Material
will NN O O
be NN O O
mapped NN O O
in NN O O
the NN O O
subspace NN O I-OtherScientificTerm
, NN O O
where NN O O
it NN O O
is NN O O
proven NN O O
that NN O O
they NN O O
form NN O O
well-defined NN O O
clusters NN O I-OtherScientificTerm
and NN O O
can NN O O
be NN O O
reliably NN O O
identified NN O O
by NN O O
a NN O O
simple NN O O
mean-shift NN O I-Method
based NN O I-Method
clustering NN O I-Method
algorithm NN O I-Method
. NN O O

Global NN O I-Task
optimality NN O I-Task
is NN O O
achieved NN O O
since NN O O
all NN O O
valid NN O O
regions NN O I-OtherScientificTerm
are NN O O
simultaneously NN O O
taken NN O O
into NN O O
account NN O O
, NN O O
and NN O O
noise NN O I-OtherScientificTerm
can NN O O
be NN O O
effectively NN O O
reduced NN O O
by NN O O
enforcing NN O O
the NN O O
subspace NN O I-OtherScientificTerm
constraint NN O I-OtherScientificTerm
. NN O O

Good NN O O
layer NN O O
descriptions NN O O
are NN O O
shown NN O O
to NN O O
be NN O O
extracted NN O O
in NN O O
the NN O O
experimental NN O O
results NN O O
. NN O O

-DOCSTART- NN O O

The NN O O
construction NN O I-Task
of NN O I-Task
causal NN O I-Task
graphs NN O I-Task
from NN O O
non-experimental NN O I-Material
data NN O I-Material
rests NN O O
on NN O O
a NN O O
set NN O O
of NN O O
constraints NN O I-OtherScientificTerm
that NN O O
the NN O O
graph NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
imposes NN O O
on NN O O
all NN O O
probability NN O I-OtherScientificTerm
distributions NN O I-OtherScientificTerm
compatible NN O O
with NN O O
the NN O O
graph NN O I-OtherScientificTerm
. NN O O

These NN O O
constraints NN O I-Generic
are NN O O
of NN O O
two NN O O
types NN O O
: NN O O
conditional NN O I-OtherScientificTerm
inde-pendencies NN O I-OtherScientificTerm
and NN O O
algebraic NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
, NN O O
first NN O O
noted NN O O
by NN O O
Verma NN O O
. NN O O

While NN O O
conditional NN O I-OtherScientificTerm
independencies NN O I-OtherScientificTerm
are NN O O
well NN O O
studied NN O O
and NN O O
frequently NN O O
used NN O O
in NN O O
causal NN O I-Method
induction NN O I-Method
algorithms NN O I-Method
, NN O O
Verma NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
are NN O O
still NN O O
poorly NN O O
understood NN O O
, NN O O
and NN O O
rarely NN O O
applied NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
examine NN O O
a NN O O
special NN O O
subset NN O O
of NN O O
Verma NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
which NN O O
are NN O O
easy NN O O
to NN O O
understand NN O O
, NN O O
easy NN O O
to NN O O
identify NN O O
and NN O O
easy NN O O
to NN O O
apply NN O O
; NN O O
they NN O I-Generic
arise NN O O
from NN O O
'' NN O O
dormant NN O I-OtherScientificTerm
independencies NN O I-OtherScientificTerm
, NN O O
'' NN O O
namely NN O O
, NN O O
conditional NN O I-OtherScientificTerm
independencies NN O I-OtherScientificTerm
that NN O O
hold NN O O
in NN O O
interventional NN O I-OtherScientificTerm
distributions NN O I-OtherScientificTerm
. NN O O

We NN O O
give NN O O
a NN O O
complete NN O O
algorithm NN O I-Generic
for NN O O
determining NN O O
if NN O O
a NN O O
dormant NN O I-OtherScientificTerm
independence NN O I-OtherScientificTerm
between NN O O
two NN O O
sets NN O O
of NN O O
variables NN O I-OtherScientificTerm
is NN O O
entailed NN O O
by NN O O
the NN O O
causal NN O I-OtherScientificTerm
graph NN O I-OtherScientificTerm
, NN O O
such NN O O
that NN O O
this NN O O
independence NN O I-OtherScientificTerm
is NN O O
identifiable NN O O
, NN O O
in NN O O
other NN O O
words NN O O
if NN O O
it NN O I-Generic
resides NN O O
in NN O O
an NN O O
interventional NN O I-OtherScientificTerm
distribution NN O I-OtherScientificTerm
that NN O O
can NN O O
be NN O O
predicted NN O O
without NN O O
resorting NN O O
to NN O O
interventions NN O I-OtherScientificTerm
. NN O O

We NN O O
further NN O O
show NN O O
the NN O O
usefulness NN O O
of NN O O
dormant NN O I-OtherScientificTerm
independencies NN O I-OtherScientificTerm
in NN O O
model NN O I-Task
testing NN O I-Task
and NN O O
induction NN O I-Task
by NN O O
giving NN O O
an NN O O
algorithm NN O I-Generic
that NN O O
uses NN O O
constraints NN O I-Generic
entailed NN O O
by NN O O
dormant NN O I-OtherScientificTerm
independencies NN O I-OtherScientificTerm
to NN O O
prune NN O O
extraneous NN O I-OtherScientificTerm
edges NN O I-OtherScientificTerm
from NN O O
a NN O O
given NN O O
causal NN O I-OtherScientificTerm
graph NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

With NN O O
the NN O O
recent NN O O
popularity NN O O
of NN O O
animated NN O I-Material
GIFs NN O I-Material
on NN O O
social NN O I-Material
media NN O I-Material
, NN O O
there NN O O
is NN O O
need NN O O
for NN O O
ways NN O O
to NN O O
index NN O O
them NN O O
with NN O O
rich NN O I-Material
meta-data NN O I-Material
. NN O O

To NN O O
advance NN O O
research NN O O
on NN O O
animated NN O I-Task
GIF NN O I-Task
understanding NN O I-Task
, NN O O
we NN O O
collected NN O O
a NN O O
new NN O O
dataset NN O I-Generic
, NN O O
Tumblr NN O I-Material
GIF NN O I-Material
-LRB- NN O I-Material
TGIF NN O I-Material
-RRB- NN O I-Material
, NN O O
with NN O O
100K NN O O
animated NN O I-Material
GIFs NN O I-Material
from NN O O
Tumblr NN O O
and NN O O
120K NN O O
natural NN O I-Material
language NN O I-Material
descriptions NN O I-Material
obtained NN O O
via NN O O
crowdsourcing NN O I-Method
. NN O O

The NN O O
motivation NN O O
for NN O O
this NN O O
work NN O O
is NN O O
to NN O O
develop NN O O
a NN O O
testbed NN O O
for NN O O
image NN O I-Task
sequence NN O I-Task
description NN O I-Task
systems NN O I-Task
, NN O O
where NN O O
the NN O O
task NN O O
is NN O O
to NN O O
generate NN O O
natural NN O I-Material
language NN O I-Material
descriptions NN O I-Material
for NN O O
animated NN O I-Material
GIFs NN O I-Material
or NN O O
video NN O I-Material
clips NN O I-Material
. NN O O

To NN O O
ensure NN O O
a NN O O
high NN O O
quality NN O O
dataset NN O O
, NN O O
we NN O O
developed NN O O
a NN O O
series NN O O
of NN O O
novel NN O O
quality NN O I-Method
controls NN O I-Method
to NN O O
validate NN O O
free-form NN O I-Material
text NN O I-Material
input NN O I-Material
from NN O O
crowd-workers NN O O
. NN O O

We NN O O
show NN O O
that NN O O
there NN O O
is NN O O
unambiguous NN O O
association NN O O
between NN O O
visual NN O I-Material
content NN O I-Material
and NN O O
natural NN O I-Material
language NN O I-Material
descriptions NN O I-Material
in NN O O
our NN O O
dataset NN O I-Generic
, NN O O
making NN O O
it NN O I-Generic
an NN O O
ideal NN O O
benchmark NN O O
for NN O O
the NN O O
visual NN O I-Task
content NN O I-Task
captioning NN O I-Task
task NN O I-Task
. NN O O

We NN O O
perform NN O O
extensive NN O O
statistical NN O I-OtherScientificTerm
analyses NN O I-OtherScientificTerm
to NN O O
compare NN O O
our NN O O
dataset NN O I-Generic
to NN O O
existing NN O O
image NN O I-Material
and NN O I-Material
video NN O I-Material
description NN O I-Material
datasets NN O I-Material
. NN O O

Next NN O O
, NN O O
we NN O O
provide NN O O
baseline NN O O
results NN O O
on NN O O
the NN O O
animated NN O I-Task
GIF NN O I-Task
description NN O I-Task
task NN O I-Task
, NN O O
using NN O O
three NN O O
representative NN O I-Generic
techniques NN O I-Generic
: NN O O
nearest NN O I-Method
neighbor NN O I-Method
, NN O O
statistical NN O I-Method
machine NN O I-Method
translation NN O I-Method
, NN O O
and NN O O
recurrent NN O I-Method
neural NN O I-Method
networks NN O I-Method
. NN O O

Finally NN O O
, NN O O
we NN O O
show NN O O
that NN O O
models NN O O
fine-tuned NN O O
from NN O O
our NN O O
animated NN O I-Material
GIF NN O I-Material
description NN O I-Material
dataset NN O I-Material
can NN O O
be NN O O
helpful NN O O
for NN O O
automatic NN O I-Task
movie NN O I-Task
description NN O I-Task
. NN O O

-DOCSTART- NN O O

Systemic NN O I-Method
grammar NN O I-Method
has NN O O
been NN O O
used NN O O
for NN O O
AI NN O I-Task
text NN O I-Task
generation NN O I-Task
work NN O O
in NN O O
the NN O O
past NN O O
, NN O O
but NN O O
the NN O O
implementations NN O I-Generic
have NN O O
tended NN O O
be NN O O
ad NN O O
hoc NN O O
or NN O O
inefficient NN O O
. NN O O

This NN O O
paper NN O O
presents NN O O
an NN O O
approach NN O I-Generic
to NN O O
systemic NN O O
text NN O I-Task
generation NN O I-Task
where NN O O
AI NN O I-Method
problem NN O I-Method
solving NN O I-Method
techniques NN O I-Method
are NN O O
applied NN O O
directly NN O O
to NN O O
an NN O O
unadulterated NN O O
systemic NN O I-Method
grammar NN O I-Method
. NN O O

This NN O O
approach NN O I-Generic
is NN O O
made NN O O
possible NN O O
by NN O O
a NN O O
special NN O O
relationship NN O O
between NN O O
systemic NN O I-Method
grammar NN O I-Method
and NN O O
problem NN O I-Method
solving NN O I-Method
: NN O O
both NN O O
are NN O O
organized NN O O
primarily NN O O
as NN O O
choosing NN O O
from NN O O
alternatives NN O O
. NN O O

The NN O O
result NN O O
is NN O O
simple NN O O
, NN O O
efficient NN O O
text NN O I-Task
generation NN O I-Task
firmly NN O O
based NN O O
in NN O O
a NN O O
linguistic NN O I-Method
theory NN O I-Method
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
a NN O O
novel NN O O
solution NN O I-Generic
to NN O O
automatic NN O I-Task
and NN O I-Task
unsupervised NN O I-Task
word NN O I-Task
sense NN O I-Task
induction NN O I-Task
-LRB- NN O I-Task
WSI NN O I-Task
-RRB- NN O I-Task
is NN O O
introduced NN O O
. NN O O

It NN O I-Generic
represents NN O O
an NN O O
instantiation NN O O
of NN O O
the NN O O
one NN O I-Method
sense NN O I-Method
per NN O I-Method
collocation NN O I-Method
observation NN O I-Method
-LRB- NN O O
Gale NN O O
et NN O O
al. NN O O
, NN O O
1992 NN O O
-RRB- NN O O
. NN O O

Like NN O O
most NN O O
existing NN O O
approaches NN O O
it NN O I-Generic
utilizes NN O O
clustering NN O I-Method
of NN O I-Method
word NN O I-Method
co-occurrences NN O I-Method
. NN O O

This NN O O
approach NN O I-Generic
differs NN O O
from NN O O
other NN O O
approaches NN O I-Generic
to NN O O
WSI NN O I-Task
in NN O O
that NN O O
it NN O I-Generic
enhances NN O O
the NN O O
effect NN O O
of NN O O
the NN O O
one NN O I-Method
sense NN O I-Method
per NN O I-Method
collocation NN O I-Method
observation NN O I-Method
by NN O O
using NN O O
triplets NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
instead NN O O
of NN O O
pairs NN O O
. NN O O

The NN O O
combination NN O O
with NN O O
a NN O O
two-step NN O I-Method
clustering NN O I-Method
process NN O I-Method
using NN O O
sentence NN O I-OtherScientificTerm
co-occurrences NN O I-OtherScientificTerm
as NN O O
features NN O I-OtherScientificTerm
allows NN O O
for NN O O
accurate NN O O
results NN O O
. NN O O

Additionally NN O O
, NN O O
a NN O O
novel NN O O
and NN O O
likewise NN O O
automatic NN O I-Metric
and NN O I-Metric
unsupervised NN O I-Metric
evaluation NN O I-Metric
method NN O I-Metric
inspired NN O O
by NN O O
Schutze NN O O
's NN O O
-LRB- NN O O
1992 NN O O
-RRB- NN O O
idea NN O O
of NN O O
evaluation NN O O
of NN O O
word NN O I-Method
sense NN O I-Method
disambiguation NN O I-Method
algorithms NN O I-Method
is NN O O
employed NN O O
. NN O O

Offering NN O O
advantages NN O O
like NN O O
reproducability NN O O
and NN O O
independency NN O O
of NN O O
a NN O O
given NN O O
biased NN O O
gold NN O O
standard NN O O
it NN O O
also NN O O
enables NN O O
automatic NN O I-Method
parameter NN O I-Method
optimization NN O I-Method
of NN O O
the NN O O
WSI NN O I-Method
algorithm NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
abstract NN O O
describes NN O O
a NN O O
natural NN O I-Method
language NN O I-Method
system NN O I-Method
which NN O O
deals NN O O
usefully NN O O
with NN O O
ungrammatical NN O I-OtherScientificTerm
input NN O I-OtherScientificTerm
and NN O O
describes NN O O
some NN O O
actual NN O O
and NN O O
potential NN O O
applications NN O O
of NN O O
it NN O I-Generic
in NN O O
computer NN O I-Task
aided NN O I-Task
second NN O I-Task
language NN O I-Task
learning NN O I-Task
. NN O O

However NN O O
, NN O O
this NN O I-Generic
is NN O O
not NN O O
the NN O O
only NN O O
area NN O O
in NN O O
which NN O O
the NN O O
principles NN O O
of NN O O
the NN O O
system NN O I-Generic
might NN O O
be NN O O
used NN O O
, NN O O
and NN O O
the NN O O
aim NN O O
in NN O O
building NN O O
it NN O I-Generic
was NN O O
simply NN O O
to NN O O
demonstrate NN O O
the NN O O
workability NN O O
of NN O O
the NN O O
general NN O O
mechanism NN O O
, NN O O
and NN O O
provide NN O O
a NN O O
framework NN O O
for NN O O
assessing NN O O
developments NN O O
of NN O O
it NN O I-Generic
. NN O O

-DOCSTART- NN O O

In NN O O
a NN O O
motorized NN O O
vehicle NN O O
a NN O O
number NN O O
of NN O O
easily NN O O
measurable NN O I-Material
signals NN O I-Material
with NN O O
frequency NN O I-Method
components NN O I-Method
related NN O O
to NN O O
the NN O O
rotational NN O I-OtherScientificTerm
speed NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
engine NN O I-OtherScientificTerm
can NN O O
be NN O O
found NN O O
, NN O O
e.g. NN O O
, NN O O
vibrations NN O I-OtherScientificTerm
, NN O O
electrical NN O I-OtherScientificTerm
system NN O I-OtherScientificTerm
voltage NN O I-OtherScientificTerm
level NN O I-OtherScientificTerm
, NN O O
and NN O O
ambient NN O I-OtherScientificTerm
sound NN O I-OtherScientificTerm
. NN O O

These NN O O
signals NN O I-Generic
could NN O O
potentially NN O O
be NN O O
used NN O O
to NN O O
estimate NN O O
the NN O O
speed NN O I-Task
and NN O I-Task
related NN O I-Task
states NN O I-Task
of NN O I-Task
the NN O I-Task
vehicle NN O I-Task
. NN O O

Unfortunately NN O O
, NN O O
such NN O O
estimates NN O O
would NN O O
typically NN O O
require NN O O
the NN O O
relations NN O O
-LRB- NN O O
scale NN O O
factors NN O O
-RRB- NN O O
between NN O O
the NN O O
frequency NN O I-Method
components NN O I-Method
and NN O O
the NN O O
speed NN O I-OtherScientificTerm
for NN O O
different NN O O
gears NN O I-OtherScientificTerm
to NN O O
be NN O O
known NN O O
. NN O O

Consequently NN O O
, NN O O
in NN O O
this NN O O
article NN O O
we NN O O
look NN O O
at NN O O
the NN O O
problem NN O O
of NN O O
estimating NN O O
these NN O O
gear NN O I-OtherScientificTerm
scale NN O I-OtherScientificTerm
factors NN O I-OtherScientificTerm
from NN O O
training NN O I-Generic
data NN O I-Generic
consisting NN O O
only NN O O
of NN O O
speed NN O I-OtherScientificTerm
measurements NN O I-OtherScientificTerm
and NN O O
measurements NN O O
of NN O O
the NN O O
signal NN O O
in NN O O
question NN O O
. NN O O

The NN O O
estimation NN O I-Task
problem NN O I-Task
is NN O O
formulated NN O O
as NN O O
a NN O O
maximum NN O I-Task
likelihood NN O I-Task
estimation NN O I-Task
problem NN O I-Task
and NN O O
heuristics NN O I-Method
is NN O O
used NN O O
to NN O O
find NN O O
initial NN O O
values NN O O
for NN O O
a NN O O
numerical NN O I-Task
evaluation NN O I-Task
of NN O I-Task
the NN O I-Task
estimator NN O I-Generic
. NN O O

Finally NN O O
, NN O O
a NN O O
measurement NN O O
campaign NN O O
is NN O O
conducted NN O O
and NN O O
the NN O O
functionality NN O O
of NN O O
the NN O O
estimation NN O I-Method
method NN O I-Method
is NN O O
verified NN O O
on NN O O
real NN O I-Material
data NN O I-Material
. NN O O

-DOCSTART- NN O O

LPC NN O I-Method
based NN O I-Method
speech NN O I-Method
coders NN O I-Method
operating NN O O
at NN O O
bit NN O I-OtherScientificTerm
rates NN O I-OtherScientificTerm
below NN O O
3.0 NN O O
kbits/sec NN O O
are NN O O
usually NN O O
associated NN O O
with NN O O
buzzy NN O I-OtherScientificTerm
or NN O I-OtherScientificTerm
metallic NN O I-OtherScientificTerm
artefacts NN O I-OtherScientificTerm
in NN O O
the NN O O
synthetic NN O I-Material
speech NN O I-Material
. NN O O

These NN O O
are NN O O
mainly NN O O
attributable NN O O
to NN O O
the NN O O
simplifying NN O O
assumptions NN O O
made NN O O
about NN O O
the NN O O
excitation NN O I-OtherScientificTerm
source NN O I-OtherScientificTerm
, NN O O
which NN O O
are NN O O
usually NN O O
required NN O O
to NN O O
maintain NN O O
such NN O O
low NN O I-OtherScientificTerm
bit NN O I-OtherScientificTerm
rates NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
paper NN O O
a NN O O
new NN O O
LPC NN O I-Method
vocoder NN O I-Method
is NN O O
presented NN O O
which NN O O
splits NN O O
the NN O O
LPC NN O I-OtherScientificTerm
excitation NN O I-OtherScientificTerm
into NN O O
two NN O O
frequency NN O I-OtherScientificTerm
bands NN O I-OtherScientificTerm
using NN O O
a NN O O
variable NN O I-OtherScientificTerm
cutoff NN O I-OtherScientificTerm
frequency NN O I-OtherScientificTerm
. NN O O

The NN O O
lower NN O O
band NN O O
is NN O O
responsible NN O O
for NN O O
representing NN O O
the NN O O
voiced NN O I-Material
parts NN O I-Material
of NN O I-Material
speech NN O I-Material
, NN O O
whilst NN O O
the NN O O
upper NN O O
band NN O O
represents NN O O
unvoiced NN O I-Material
speech NN O I-Material
. NN O O

In NN O O
doing NN O O
so NN O O
the NN O O
coder NN O I-Generic
's NN O O
performance NN O O
during NN O O
both NN O O
mixed NN O I-Material
voicing NN O I-Material
speech NN O I-Material
and NN O O
speech NN O I-Material
containing NN O I-Material
acoustic NN O I-Material
noise NN O I-Material
is NN O O
greatly NN O O
improved NN O O
, NN O O
producing NN O O
soft NN O I-Material
natural NN O I-Material
sounding NN O I-Material
speech NN O I-Material
. NN O O

The NN O O
paper NN O O
also NN O O
describes NN O O
new NN O O
parameter NN O I-Method
determination NN O I-Method
and NN O O
quantisation NN O I-Method
techniques NN O I-Method
vital NN O O
to NN O O
the NN O O
operation NN O O
of NN O O
this NN O O
coder NN O I-Generic
at NN O O
such NN O O
low NN O I-OtherScientificTerm
bit NN O I-OtherScientificTerm
rates NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

We NN O O
consider NN O O
a NN O O
problem NN O O
of NN O O
blind NN O I-Task
source NN O I-Task
separation NN O I-Task
from NN O O
a NN O O
set NN O O
of NN O O
instantaneous NN O I-OtherScientificTerm
linear NN O I-OtherScientificTerm
mixtures NN O I-OtherScientificTerm
, NN O O
where NN O O
the NN O O
mixing NN O I-OtherScientificTerm
matrix NN O I-OtherScientificTerm
is NN O O
unknown NN O O
. NN O O

It NN O O
was NN O O
discovered NN O O
recently NN O O
, NN O O
that NN O O
exploiting NN O O
the NN O O
sparsity NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
sources NN O I-OtherScientificTerm
in NN O O
an NN O O
appropriate NN O O
representation NN O O
according NN O O
to NN O O
some NN O O
signal NN O I-OtherScientificTerm
dictionary NN O I-OtherScientificTerm
, NN O O
dramatically NN O O
improves NN O O
the NN O O
quality NN O I-Metric
of NN O I-Metric
separation NN O I-Metric
. NN O O

In NN O O
this NN O O
work NN O O
we NN O O
use NN O O
the NN O O
property NN O O
of NN O O
multi NN O I-OtherScientificTerm
scale NN O I-OtherScientificTerm
transforms NN O I-OtherScientificTerm
, NN O O
such NN O O
as NN O O
wavelet NN O I-OtherScientificTerm
or NN O I-OtherScientificTerm
wavelet NN O I-OtherScientificTerm
packets NN O I-OtherScientificTerm
, NN O O
to NN O O
decompose NN O O
signals NN O O
into NN O O
sets NN O O
of NN O O
local NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
with NN O O
various NN O O
degrees NN O O
of NN O O
sparsity NN O I-OtherScientificTerm
. NN O O

We NN O O
use NN O O
this NN O O
intrinsic NN O O
property NN O O
for NN O O
selecting NN O O
the NN O O
best NN O O
-LRB- NN O O
most NN O O
sparse NN O O
-RRB- NN O O
subsets NN O O
of NN O O
features NN O I-OtherScientificTerm
for NN O O
further NN O O
separation NN O O
. NN O O

The NN O O
performance NN O O
of NN O O
the NN O O
algorithm NN O I-Generic
is NN O O
verified NN O O
on NN O O
noise-free NN O I-Material
and NN O I-Material
noisy NN O I-Material
data NN O I-Material
. NN O O

Experiments NN O O
with NN O O
simulated NN O I-Material
signals NN O I-Material
, NN O O
musical NN O I-Material
sounds NN O I-Material
and NN O O
images NN O I-Material
demonstrate NN O O
significant NN O O
improvement NN O O
of NN O O
separation NN O I-Metric
quality NN O I-Metric
over NN O O
previously NN O O
reported NN O O
results NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
explore NN O O
multilingual NN O I-Task
feature-level NN O I-Task
data NN O I-Task
sharing NN O I-Task
via NN O O
Deep NN O I-Method
Neural NN O I-Method
Network NN O I-Method
-LRB- NN O I-Method
DNN NN O I-Method
-RRB- NN O I-Method
stacked NN O I-Method
bottleneck NN O I-Method
features NN O I-Method
. NN O O

Given NN O O
a NN O O
set NN O O
of NN O O
available NN O O
source NN O O
languages NN O O
, NN O O
we NN O O
apply NN O O
language NN O I-Task
identification NN O I-Task
to NN O O
pick NN O O
the NN O O
language NN O O
most NN O O
similar NN O O
to NN O O
the NN O O
target NN O O
language NN O O
, NN O O
for NN O O
more NN O O
efficient NN O O
use NN O O
of NN O O
multilingual NN O I-Material
resources NN O I-Material
. NN O O

Our NN O O
experiments NN O O
with NN O O
IARPA-Babel NN O I-Material
languages NN O I-Material
show NN O O
that NN O O
bottleneck NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
trained NN O O
on NN O O
the NN O O
most NN O O
similar NN O O
source NN O O
language NN O O
perform NN O O
better NN O O
than NN O O
those NN O I-Generic
trained NN O O
on NN O O
all NN O O
available NN O O
source NN O O
languages NN O O
. NN O O

Further NN O O
analysis NN O O
suggests NN O O
that NN O O
only NN O O
data NN O I-Generic
similar NN O O
to NN O O
the NN O O
target NN O O
language NN O O
is NN O O
useful NN O O
for NN O O
multilingual NN O I-Task
training NN O I-Task
. NN O O

-DOCSTART- NN O O

This NN O O
article NN O O
introduces NN O O
a NN O O
bidirectional NN O I-Method
grammar NN O I-Method
generation NN O I-Method
system NN O I-Method
called NN O O
feature NN O I-Method
structure-directed NN O I-Method
generation NN O I-Method
, NN O O
developed NN O O
for NN O O
a NN O O
dialogue NN O I-Task
translation NN O I-Task
system NN O I-Task
. NN O O

The NN O O
system NN O I-Generic
utilizes NN O O
typed NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
to NN O O
control NN O O
the NN O O
top-down NN O I-OtherScientificTerm
derivation NN O I-OtherScientificTerm
in NN O O
a NN O O
declarative NN O O
way NN O O
. NN O O

This NN O O
generation NN O I-Method
system NN O I-Method
also NN O O
uses NN O O
disjunctive NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
to NN O O
reduce NN O O
the NN O O
number NN O O
of NN O O
copies NN O O
of NN O O
the NN O O
derivation NN O I-OtherScientificTerm
tree NN O I-OtherScientificTerm
. NN O O

The NN O O
grammar NN O I-Method
for NN O O
this NN O O
generator NN O I-Generic
is NN O O
designed NN O O
to NN O O
properly NN O O
generate NN O O
the NN O O
speaker NN O I-OtherScientificTerm
's NN O I-OtherScientificTerm
intention NN O I-OtherScientificTerm
in NN O O
a NN O O
telephone NN O I-OtherScientificTerm
dialogue NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Automatic NN O I-Task
image NN O I-Task
annotation NN O I-Task
is NN O O
a NN O O
newly NN O O
developed NN O O
and NN O O
promising NN O O
technique NN O O
to NN O O
provide NN O O
semantic NN O I-Task
image NN O I-Task
retrieval NN O I-Task
via NN O O
text NN O I-Material
descriptions NN O I-Material
. NN O O

It NN O O
concerns NN O O
a NN O O
process NN O O
of NN O O
automatically NN O I-Task
labeling NN O I-Task
the NN O I-Task
image NN O I-Task
contents NN O I-Task
with NN O O
a NN O O
pre-defined NN O O
set NN O O
of NN O O
keywords NN O I-OtherScientificTerm
which NN O O
are NN O O
exploited NN O O
to NN O O
represent NN O O
the NN O O
image NN O I-OtherScientificTerm
semantics NN O I-OtherScientificTerm
. NN O O

A NN O O
Maximum NN O I-Method
Entropy NN O I-Method
Model-based NN O I-Method
approach NN O I-Method
to NN O O
the NN O O
task NN O O
of NN O O
automatic NN O I-Task
image NN O I-Task
annotation NN O I-Task
is NN O O
proposed NN O O
in NN O O
this NN O O
paper NN O O
. NN O O

In NN O O
the NN O O
phase NN O O
of NN O O
training NN O I-Task
, NN O O
a NN O O
basic NN O O
visual NN O I-OtherScientificTerm
vocabulary NN O I-OtherScientificTerm
consisting NN O O
of NN O O
blob-tokens NN O I-OtherScientificTerm
to NN O O
describe NN O O
the NN O O
image NN O I-OtherScientificTerm
content NN O I-OtherScientificTerm
is NN O O
generated NN O O
at NN O O
first NN O O
; NN O O
then NN O O
the NN O O
statistical NN O I-Generic
relationship NN O I-Generic
is NN O O
modeled NN O O
between NN O O
the NN O O
blob-tokens NN O I-OtherScientificTerm
and NN O O
keywords NN O I-OtherScientificTerm
by NN O O
a NN O O
Maximum NN O I-Method
Entropy NN O I-Method
Model NN O I-Method
constructed NN O O
from NN O O
the NN O O
training NN O O
set NN O O
of NN O O
labeled NN O O
images NN O O
. NN O O

In NN O O
the NN O O
phase NN O O
of NN O O
annotation NN O I-Task
, NN O O
for NN O O
an NN O O
unlabeled NN O O
image NN O O
, NN O O
the NN O O
most NN O O
likely NN O O
associated NN O O
keywords NN O I-OtherScientificTerm
are NN O O
predicted NN O O
in NN O O
terms NN O O
of NN O O
the NN O O
blob-token NN O I-OtherScientificTerm
set NN O I-OtherScientificTerm
extracted NN O O
from NN O O
the NN O O
given NN O O
image NN O O
. NN O O

We NN O O
carried NN O O
out NN O O
experiments NN O O
on NN O O
a NN O O
medium-sized NN O I-Material
image NN O I-Material
collection NN O I-Material
with NN O O
about NN O O
5000 NN O O
images NN O O
from NN O O
Corel NN O I-Material
Photo NN O I-Material
CDs NN O I-Material
. NN O O

The NN O O
experimental NN O O
results NN O O
demonstrated NN O O
that NN O O
the NN O O
annotation NN O I-Task
performance NN O O
of NN O O
this NN O O
method NN O I-Generic
outperforms NN O O
some NN O O
traditional NN O O
annotation NN O I-Method
methods NN O I-Method
by NN O O
about NN O O
8 NN O O
% NN O O
in NN O O
mean NN O I-Metric
precision NN O I-Metric
, NN O O
showing NN O O
a NN O O
potential NN O O
of NN O O
the NN O O
Maximum NN O I-Method
Entropy NN O I-Method
Model NN O I-Method
in NN O O
the NN O O
task NN O O
of NN O O
automatic NN O I-Task
image NN O I-Task
annotation NN O I-Task
. NN O O

-DOCSTART- NN O O

Recent NN O O
years NN O O
have NN O O
seen NN O O
increasing NN O O
research NN O O
on NN O O
extracting NN O O
and NN O O
using NN O O
temporal NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
in NN O O
natural NN O I-Task
language NN O I-Task
applications NN O I-Task
. NN O O

However NN O O
most NN O O
of NN O O
the NN O O
works NN O O
found NN O O
in NN O O
the NN O O
literature NN O O
have NN O O
focused NN O O
on NN O O
identifying NN O O
and NN O O
understanding NN O O
temporal NN O I-OtherScientificTerm
expressions NN O I-OtherScientificTerm
in NN O I-OtherScientificTerm
newswire NN O I-Material
texts NN O I-Material
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
report NN O O
our NN O O
work NN O O
on NN O O
anchoring NN O O
temporal NN O I-OtherScientificTerm
expressions NN O I-OtherScientificTerm
in NN O O
a NN O O
novel NN O O
genre NN O O
, NN O O
emails NN O I-Material
. NN O O

The NN O O
highly NN O O
under-specified NN O O
nature NN O O
of NN O O
these NN O O
expressions NN O I-Generic
fits NN O O
well NN O O
with NN O O
our NN O O
constraint-based NN O I-OtherScientificTerm
representation NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
time NN O I-OtherScientificTerm
, NN O O
Time NN O I-OtherScientificTerm
Calculus NN O I-OtherScientificTerm
for NN O I-OtherScientificTerm
Natural NN O I-OtherScientificTerm
Language NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
TCNL NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
. NN O O

We NN O O
have NN O O
developed NN O O
and NN O O
evaluated NN O O
a NN O O
Temporal NN O I-Method
Expression NN O I-Method
Anchoror NN O I-Method
-LRB- NN O I-Method
TEA NN O I-Method
-RRB- NN O I-Method
, NN O O
and NN O O
the NN O O
result NN O O
shows NN O O
that NN O O
it NN O I-Generic
performs NN O O
significantly NN O O
better NN O O
than NN O O
the NN O O
baseline NN O I-Generic
, NN O O
and NN O O
compares NN O O
favorably NN O O
with NN O O
some NN O O
of NN O O
the NN O O
closely NN O O
related NN O O
work NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
address NN O O
the NN O O
problem NN O O
of NN O O
populating NN O O
object NN O I-Material
category NN O I-Material
detection NN O I-Material
datasets NN O I-Material
with NN O O
dense NN O O
, NN O O
per-object NN O I-Task
3D NN O I-Task
reconstructions NN O I-Task
, NN O O
bootstrapped NN O O
from NN O O
class NN O O
labels NN O O
, NN O O
ground NN O I-Method
truth NN O I-Method
figure-ground NN O I-Method
segmentations NN O I-Method
and NN O O
a NN O O
small NN O O
set NN O O
of NN O O
keypoint NN O I-Method
annotations NN O I-Method
. NN O O

Our NN O O
proposed NN O O
algorithm NN O I-Generic
first NN O O
estimates NN O O
camera NN O I-OtherScientificTerm
viewpoint NN O I-OtherScientificTerm
using NN O O
rigid NN O I-OtherScientificTerm
structure-from-motion NN O I-OtherScientificTerm
, NN O O
then NN O O
reconstructs NN O O
object NN O I-OtherScientificTerm
shapes NN O I-OtherScientificTerm
by NN O O
optimizing NN O O
over NN O O
visual NN O I-OtherScientificTerm
hull NN O I-OtherScientificTerm
proposals NN O I-OtherScientificTerm
guided NN O O
by NN O O
loose NN O I-OtherScientificTerm
within-class NN O I-OtherScientificTerm
shape NN O I-OtherScientificTerm
similarity NN O I-OtherScientificTerm
assumptions NN O I-OtherScientificTerm
. NN O O

The NN O O
visual NN O I-Method
hull NN O I-Method
sampling NN O I-Method
process NN O I-Method
attempts NN O O
to NN O O
intersect NN O O
an NN O O
object NN O O
's NN O O
projection NN O I-OtherScientificTerm
cone NN O I-OtherScientificTerm
with NN O O
the NN O O
cones NN O O
of NN O O
minimal NN O O
subsets NN O O
of NN O O
other NN O O
similar NN O O
objects NN O O
among NN O O
those NN O O
pictured NN O O
from NN O O
certain NN O O
vantage NN O O
points NN O O
. NN O O

We NN O O
show NN O O
that NN O O
our NN O O
method NN O I-Generic
is NN O O
able NN O O
to NN O O
produce NN O O
convincing NN O O
per-object NN O I-Task
3D NN O I-Task
reconstructions NN O I-Task
on NN O O
one NN O O
of NN O O
the NN O O
most NN O O
challenging NN O O
existing NN O O
object-category NN O I-Material
detection NN O I-Material
datasets NN O I-Material
, NN O O
PASCAL NN O I-Material
VOC NN O I-Material
. NN O O

Our NN O O
results NN O O
may NN O O
re-stimulate NN O O
once NN O O
popular NN O O
geometry-oriented NN O I-Method
model-based NN O I-Method
recognition NN O I-Method
approaches NN O I-Method
. NN O O

-DOCSTART- NN O O

Probabilistic NN O I-Method
models NN O I-Method
have NN O O
been NN O O
previously NN O O
shown NN O O
to NN O O
be NN O O
efficient NN O O
and NN O O
effective NN O O
for NN O O
modeling NN O I-Task
and NN O I-Task
recognition NN O I-Task
of NN O I-Task
human NN O I-Task
motion NN O I-Task
. NN O O

In NN O O
particular NN O O
we NN O O
focus NN O O
on NN O O
methods NN O O
which NN O O
represent NN O O
the NN O O
human NN O I-Method
motion NN O I-Method
model NN O I-Method
as NN O O
a NN O O
triangulated NN O I-Method
graph NN O I-Method
. NN O O

Previous NN O O
approaches NN O O
learned NN O O
models NN O I-Generic
based NN O O
just NN O O
on NN O O
positions NN O I-OtherScientificTerm
and NN O O
velocities NN O I-OtherScientificTerm
of NN O O
the NN O O
body NN O O
parts NN O O
while NN O O
ignoring NN O O
their NN O O
appearance NN O I-OtherScientificTerm
. NN O O

Moreover NN O O
, NN O O
a NN O O
heuristic NN O I-Method
approach NN O I-Method
was NN O O
commonly NN O O
used NN O O
to NN O O
obtain NN O O
translation NN O I-OtherScientificTerm
invariance NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
suggest NN O O
an NN O O
improved NN O O
approach NN O I-Generic
for NN O O
learning NN O O
such NN O O
models NN O I-Generic
and NN O O
using NN O O
them NN O I-Generic
for NN O O
human NN O I-Task
motion NN O I-Task
recognition NN O I-Task
. NN O O

The NN O O
suggested NN O O
approach NN O I-Generic
combines NN O O
multiple NN O O
cues NN O I-Generic
, NN O O
i.e. NN O O
, NN O O
positions NN O I-OtherScientificTerm
, NN O O
velocities NN O I-OtherScientificTerm
and NN O O
appearance NN O I-OtherScientificTerm
into NN O O
both NN O O
the NN O O
learning NN O I-Task
and NN O I-Task
detection NN O I-Task
phases NN O I-Task
. NN O O

Furthermore NN O O
, NN O O
we NN O O
introduce NN O O
global NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
in NN O O
the NN O O
model NN O I-Generic
, NN O O
which NN O O
can NN O O
represent NN O O
global NN O I-OtherScientificTerm
properties NN O I-OtherScientificTerm
such NN O O
as NN O O
translation NN O I-OtherScientificTerm
, NN O O
scale NN O I-OtherScientificTerm
or NN O O
viewpoint NN O I-OtherScientificTerm
. NN O O

The NN O O
model NN O I-Generic
is NN O O
learned NN O O
in NN O O
an NN O O
unsupervised NN O I-Method
manner NN O I-Method
from NN O O
un-labelled NN O I-Material
data NN O I-Material
. NN O O

We NN O O
show NN O O
that NN O O
the NN O O
suggested NN O O
hybrid NN O I-Method
proba-bilistic NN O I-Method
model NN O I-Method
-LRB- NN O O
which NN O O
combines NN O O
global NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
, NN O O
like NN O O
translation NN O I-OtherScientificTerm
, NN O O
with NN O O
local NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
, NN O O
like NN O O
relative NN O I-OtherScientificTerm
positions NN O I-OtherScientificTerm
and NN O O
appearances NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
body NN O I-OtherScientificTerm
parts NN O I-OtherScientificTerm
-RRB- NN O O
, NN O O
leads NN O O
to NN O O
: NN O O
-LRB- NN O O
i NN O O
-RRB- NN O O
faster NN O I-Metric
convergence NN O I-Metric
of NN O O
learning NN O I-Task
phase NN O I-Task
, NN O O
-LRB- NN O O
ii NN O O
-RRB- NN O O
robustness NN O I-Metric
to NN O O
occlusions NN O I-OtherScientificTerm
, NN O O
and NN O O
, NN O O
-LRB- NN O O
iii NN O O
-RRB- NN O O
higher NN O O
recognition NN O I-Material
rate NN O I-Material
. NN O O

-DOCSTART- NN O O

Factor NN O I-Method
analysis NN O I-Method
and NN O O
principal NN O I-Method
components NN O I-Method
analysis NN O I-Method
can NN O O
be NN O O
used NN O O
to NN O O
model NN O O
linear NN O I-OtherScientificTerm
relationships NN O I-OtherScientificTerm
between NN O I-OtherScientificTerm
observed NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
and NN O O
linearly NN O O
map NN O O
high-dimensional NN O I-Material
data NN O I-Material
to NN O O
a NN O O
lower-dimensional NN O I-OtherScientificTerm
hidden NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
. NN O O

In NN O O
factor NN O I-Method
analysis NN O I-Method
, NN O O
the NN O O
observations NN O O
are NN O O
modeled NN O O
as NN O O
a NN O O
linear NN O I-Method
combination NN O I-Method
of NN O I-Method
normally NN O I-Method
distributed NN O I-Method
hidden NN O I-Method
variables NN O I-Method
. NN O O

We NN O O
describe NN O O
a NN O O
nonlinear NN O I-Method
generalization NN O I-Method
of NN O I-Method
factor NN O I-Method
analysis NN O I-Method
, NN O O
called NN O O
`` NN O I-Method
product NN O I-Method
analy-sis NN O I-Method
'' NN O I-Method
, NN O O
that NN O O
models NN O O
the NN O O
observed NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
as NN O O
a NN O O
linear NN O I-OtherScientificTerm
combination NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
products NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
normally NN O I-OtherScientificTerm
distributed NN O I-OtherScientificTerm
hidden NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
. NN O O

Just NN O O
as NN O O
factor NN O I-Method
analysis NN O I-Method
can NN O O
be NN O O
viewed NN O O
as NN O O
unsupervised NN O I-Method
linear NN O I-Method
regression NN O I-Method
on NN O O
unobserved NN O O
, NN O O
normally NN O O
distributed NN O I-OtherScientificTerm
hidden NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
, NN O O
product NN O I-Method
analysis NN O I-Method
can NN O O
be NN O O
viewed NN O O
as NN O O
unsupervised NN O I-Method
linear NN O I-Method
regression NN O I-Method
on NN O O
products NN O O
of NN O O
unobserved NN O O
, NN O O
normally NN O O
distributed NN O I-OtherScientificTerm
hidden NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
. NN O O

The NN O O
mapping NN O O
between NN O O
the NN O O
data NN O O
and NN O O
the NN O O
hidden NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
is NN O O
nonlinear NN O O
, NN O O
so NN O O
we NN O O
use NN O O
an NN O O
approximate NN O I-Method
variational NN O I-Method
technique NN O I-Method
for NN O O
inference NN O I-Task
and NN O O
learning NN O I-Task
. NN O O

Since NN O O
product NN O I-Method
analysis NN O I-Method
is NN O O
a NN O O
generalization NN O I-Method
of NN O I-Method
factor NN O I-Method
analysis NN O I-Method
, NN O O
product NN O I-Method
analysis NN O I-Method
always NN O O
finds NN O O
a NN O O
higher NN O O
data NN O O
likelihood NN O O
than NN O O
factor NN O I-Method
analysis NN O I-Method
. NN O O

We NN O O
give NN O O
results NN O O
on NN O O
pattern NN O I-Task
recognition NN O I-Task
and NN O O
illumination-invariant NN O I-Task
image NN O I-Task
clustering NN O I-Task
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
a NN O O
domain NN O I-Method
independent NN O I-Method
strategy NN O I-Method
for NN O O
the NN O O
multimedia NN O I-Task
articulation NN O I-Task
of NN O I-Task
answers NN O I-Task
elicited NN O O
by NN O O
a NN O O
natural NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
interface NN O I-OtherScientificTerm
to NN O O
database NN O I-Task
query NN O I-Task
applications NN O I-Task
. NN O O

Multimedia NN O I-Material
answers NN O I-Material
include NN O O
videodisc NN O I-Material
images NN O I-Material
and NN O O
heuristically-produced NN O O
complete NN O O
sentences NN O O
in NN O O
text NN O O
or NN O O
text-to-speech NN O I-OtherScientificTerm
form NN O I-OtherScientificTerm
. NN O O

Deictic NN O I-OtherScientificTerm
reference NN O I-OtherScientificTerm
and NN O O
feedback NN O I-OtherScientificTerm
about NN O O
the NN O O
discourse NN O I-OtherScientificTerm
are NN O O
enabled NN O O
. NN O O

The NN O O
interface NN O I-Generic
thus NN O O
presents NN O O
the NN O O
application NN O I-Generic
as NN O O
cooperative NN O O
and NN O O
conversational NN O O
. NN O O

-DOCSTART- NN O O

The NN O O
LOGON NN O I-Method
MT NN O I-Method
demonstrator NN O I-Method
assembles NN O O
independently NN O O
valuable NN O O
general-purpose NN O I-Method
NLP NN O I-Method
components NN O I-Method
into NN O O
a NN O O
machine NN O I-Method
translation NN O I-Method
pipeline NN O I-Method
that NN O O
capitalizes NN O O
on NN O O
output NN O O
quality NN O O
. NN O O

The NN O O
demonstrator NN O I-Task
embodies NN O O
an NN O O
interesting NN O O
combination NN O O
of NN O O
hand-built NN O I-Material
, NN O I-Material
symbolic NN O I-Material
resources NN O I-Material
and NN O O
stochastic NN O I-Method
processes NN O I-Method
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
investigate NN O O
the NN O O
simple NN O O
logical NN O I-Task
properties NN O I-Task
of NN O I-Task
contexts NN O I-Task
. NN O O

We NN O O
describe NN O O
both NN O O
the NN O O
syntax NN O I-OtherScientificTerm
and NN O O
semantics NN O I-OtherScientificTerm
of NN O O
a NN O O
general NN O O
propositional NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
context NN O I-OtherScientificTerm
, NN O O
and NN O O
give NN O O
a NN O O
Hilbert NN O I-Method
style NN O I-Method
proof NN O I-Method
system NN O I-Method
for NN O O
this NN O O
language NN O I-Generic
. NN O O

A NN O O
propositional NN O I-OtherScientificTerm
logic NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
context NN O I-OtherScientificTerm
extends NN O O
classical NN O I-OtherScientificTerm
propositional NN O I-OtherScientificTerm
logic NN O I-OtherScientificTerm
in NN O O
two NN O O
ways NN O O
. NN O O

Firstly NN O O
, NN O O
a NN O O
new NN O O
modality NN O I-OtherScientificTerm
, NN O O
ist NN O O
-LRB- NN O O
;-RRB- NN O O
, NN O O
is NN O O
introduced NN O O
. NN O O

It NN O I-Generic
is NN O O
used NN O O
to NN O O
express NN O O
that NN O O
the NN O O
sentence NN O O
, NN O O
, NN O O
holds NN O O
in NN O O
the NN O O
context NN O O
. NN O O

Secondly NN O O
, NN O O
each NN O O
context NN O O
has NN O O
its NN O O
own NN O O
vocabulary NN O O
, NN O O
i.e. NN O O
a NN O O
set NN O O
of NN O O
propositional NN O O
atoms NN O O
which NN O O
are NN O O
deened NN O O
or NN O O
meaningful NN O O
in NN O O
that NN O O
context NN O O
. NN O O

The NN O O
main NN O O
results NN O O
of NN O O
this NN O O
paper NN O O
are NN O O
the NN O O
sound-ness NN O O
and NN O O
completeness NN O O
of NN O O
this NN O O
Hilbert NN O I-Method
style NN O I-Method
proof NN O I-Method
system NN O I-Method
. NN O O

We NN O O
also NN O O
provide NN O O
soundness NN O O
and NN O O
completeness NN O O
results NN O O
-LRB- NN O O
i.e. NN O O
correspondence NN O I-OtherScientificTerm
theory NN O I-OtherScientificTerm
-RRB- NN O O
for NN O O
various NN O O
extensions NN O O
of NN O O
the NN O O
general NN O O
system NN O O
. NN O O

-DOCSTART- NN O O

Image NN O I-Task
matching NN O I-Task
is NN O O
a NN O O
fundamental NN O O
problem NN O O
in NN O O
Computer NN O I-Task
Vision NN O I-Task
. NN O O

In NN O O
the NN O O
context NN O O
of NN O O
feature-based NN O I-Task
matching NN O I-Task
, NN O O
SIFT NN O I-Method
and NN O O
its NN O O
variants NN O O
have NN O O
long NN O O
excelled NN O O
in NN O O
a NN O O
wide NN O O
array NN O O
of NN O O
applications NN O O
. NN O O

However NN O O
, NN O O
for NN O O
ultra-wide NN O I-OtherScientificTerm
baselines NN O I-OtherScientificTerm
, NN O O
as NN O O
in NN O O
the NN O O
case NN O O
of NN O O
aerial NN O I-Material
images NN O I-Material
captured NN O O
under NN O O
large NN O I-OtherScientificTerm
camera NN O I-OtherScientificTerm
rotations NN O I-OtherScientificTerm
, NN O O
the NN O O
appearance NN O I-OtherScientificTerm
variation NN O I-OtherScientificTerm
goes NN O O
beyond NN O O
the NN O O
reach NN O O
of NN O O
SIFT NN O I-Method
and NN O O
RANSAC NN O I-Method
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
propose NN O O
a NN O O
data-driven NN O I-Method
, NN O I-Method
deep NN O I-Method
learning-based NN O I-Method
approach NN O I-Method
that NN O O
sidesteps NN O O
local NN O I-OtherScientificTerm
correspondence NN O I-OtherScientificTerm
by NN O O
framing NN O O
the NN O O
problem NN O I-Generic
as NN O O
a NN O O
classification NN O I-Task
task NN O I-Task
. NN O O

Furthermore NN O O
, NN O O
we NN O O
demonstrate NN O O
that NN O O
local NN O I-OtherScientificTerm
correspondences NN O I-OtherScientificTerm
can NN O O
still NN O O
be NN O O
useful NN O O
. NN O O

To NN O O
do NN O O
so NN O O
we NN O O
incorporate NN O O
an NN O O
attention NN O I-Method
mechanism NN O I-Method
to NN O O
produce NN O O
a NN O O
set NN O O
of NN O O
probable NN O O
matches NN O O
, NN O O
which NN O O
allows NN O O
us NN O O
to NN O O
further NN O O
increase NN O O
performance NN O O
. NN O O

We NN O O
train NN O O
our NN O O
models NN O I-Generic
on NN O O
a NN O O
dataset NN O I-Material
of NN O I-Material
urban NN O I-Material
aerial NN O I-Material
imagery NN O I-Material
consisting NN O O
of NN O O
` NN O O
same NN O O
' NN O O
and NN O O
` NN O O
different NN O O
' NN O O
pairs NN O O
, NN O O
collected NN O O
for NN O O
this NN O O
purpose NN O O
, NN O O
and NN O O
characterize NN O O
the NN O O
problem NN O I-Generic
via NN O O
a NN O O
human NN O I-Method
study NN O I-Method
with NN O O
annotations NN O I-Material
from NN O I-Material
Amazon NN O I-Material
Mechanical NN O I-Material
Turk NN O I-Material
. NN O O

We NN O O
demonstrate NN O O
that NN O O
our NN O O
models NN O I-Generic
outperform NN O O
the NN O O
state-of-the-art NN O I-Generic
on NN O O
ultra-wide NN O I-Task
baseline NN O I-Task
matching NN O I-Task
and NN O O
approach NN O O
human NN O I-Metric
accuracy NN O I-Metric
. NN O O

-DOCSTART- NN O O

This NN O O
article NN O O
is NN O O
devoted NN O O
to NN O O
the NN O O
problem NN O O
of NN O O
quantifying NN O I-Task
noun NN O I-Task
groups NN O I-Task
in NN O I-Task
German NN O I-Material
. NN O O

After NN O O
a NN O O
thorough NN O O
description NN O O
of NN O O
the NN O O
phenomena NN O O
, NN O O
the NN O O
results NN O O
of NN O O
corpus-based NN O O
investigations NN O O
are NN O O
described NN O O
. NN O O

Moreover NN O O
, NN O O
some NN O O
examples NN O O
are NN O O
given NN O O
that NN O O
underline NN O O
the NN O O
necessity NN O O
of NN O O
integrating NN O O
some NN O O
kind NN O O
of NN O O
information NN O O
other NN O O
than NN O O
grammar NN O I-OtherScientificTerm
sensu NN O I-OtherScientificTerm
stricto NN O I-OtherScientificTerm
into NN O O
the NN O O
treebank NN O I-Material
. NN O O

We NN O O
argue NN O O
that NN O O
a NN O O
more NN O O
sophisticated NN O O
and NN O O
fine-grained NN O I-OtherScientificTerm
annotation NN O I-OtherScientificTerm
in NN O O
the NN O O
tree-bank NN O I-Material
would NN O O
have NN O O
very NN O O
positve NN O O
effects NN O O
on NN O O
stochastic NN O I-Method
parsers NN O I-Method
trained NN O O
on NN O O
the NN O O
tree-bank NN O I-Material
and NN O O
on NN O O
grammars NN O I-Method
induced NN O O
from NN O O
the NN O O
treebank NN O I-Material
, NN O O
and NN O O
it NN O O
would NN O O
make NN O O
the NN O O
treebank NN O I-Material
more NN O O
valuable NN O O
as NN O O
a NN O O
source NN O O
of NN O O
data NN O O
for NN O O
theoretical NN O I-Task
linguistic NN O I-Task
investigations NN O I-Task
. NN O O

The NN O O
information NN O O
gained NN O O
from NN O O
corpus NN O O
research NN O O
and NN O O
the NN O O
analyses NN O O
that NN O O
are NN O O
proposed NN O O
are NN O O
realized NN O O
in NN O O
the NN O O
framework NN O O
of NN O O
SILVA NN O I-Method
, NN O O
a NN O O
parsing NN O I-Method
and NN O I-Method
extraction NN O I-Method
tool NN O I-Method
for NN O O
German NN O I-Material
text NN O I-Material
corpora NN O I-Material
. NN O O

-DOCSTART- NN O O

While NN O O
paraphrasing NN O I-Method
is NN O O
critical NN O O
both NN O O
for NN O O
interpretation NN O I-Task
and NN O I-Task
generation NN O I-Task
of NN O I-Task
natural NN O I-Task
language NN O I-Task
, NN O O
current NN O O
systems NN O I-Generic
use NN O O
manual NN O I-Method
or NN O I-Method
semi-automatic NN O I-Method
methods NN O I-Method
to NN O O
collect NN O O
paraphrases NN O I-OtherScientificTerm
. NN O O

We NN O O
present NN O O
an NN O O
unsupervised NN O I-Method
learning NN O I-Method
algorithm NN O I-Method
for NN O O
identification NN O I-Task
of NN O I-Task
paraphrases NN O I-Task
from NN O O
a NN O O
corpus NN O I-Material
of NN O I-Material
multiple NN O I-Material
English NN O I-Material
translations NN O I-Material
of NN O O
the NN O O
same NN O O
source NN O O
text NN O O
. NN O O

Our NN O O
approach NN O I-Generic
yields NN O O
phrasal NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
single NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
lexical NN O I-OtherScientificTerm
paraphrases NN O I-OtherScientificTerm
as NN O O
well NN O O
as NN O O
syntactic NN O I-OtherScientificTerm
paraphrases NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

An NN O O
efficient NN O O
bit-vector-based NN O I-Method
CKY-style NN O I-Method
parser NN O I-Method
for NN O O
context-free NN O I-Task
parsing NN O I-Task
is NN O O
presented NN O O
. NN O O

The NN O O
parser NN O I-Method
computes NN O O
a NN O O
compact NN O O
parse NN O I-Method
forest NN O I-Method
representation NN O I-Method
of NN O O
the NN O O
complete NN O O
set NN O O
of NN O O
possible NN O O
analyses NN O O
for NN O O
large NN O I-Method
treebank NN O I-Method
grammars NN O I-Method
and NN O O
long NN O O
input NN O O
sentences NN O O
. NN O O

The NN O O
parser NN O I-Method
uses NN O O
bit-vector NN O I-Method
operations NN O I-Method
to NN O O
parallelise NN O O
the NN O O
basic NN O O
parsing NN O O
operations NN O O
. NN O O

The NN O O
parser NN O I-Method
is NN O O
particularly NN O O
useful NN O O
when NN O O
all NN O O
analyses NN O O
are NN O O
needed NN O O
rather NN O O
than NN O O
just NN O O
the NN O O
most NN O O
probable NN O O
one NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
partially-blurred-image NN O I-Method
classification NN O I-Method
and NN O I-Method
analysis NN O I-Method
framework NN O I-Method
for NN O O
automatically NN O I-Task
detecting NN O I-Task
images NN O I-Material
containing NN O O
blurred NN O I-OtherScientificTerm
regions NN O I-OtherScientificTerm
and NN O O
recognizing NN O O
the NN O O
blur NN O O
types NN O O
for NN O O
those NN O O
regions NN O I-Generic
without NN O O
needing NN O O
to NN O O
perform NN O O
blur NN O I-Task
kernel NN O I-Task
estimation NN O I-Task
and NN O O
image NN O I-Task
deblurring NN O I-Task
. NN O O

We NN O O
develop NN O O
several NN O O
blur NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
modeled NN O O
by NN O O
image NN O I-OtherScientificTerm
color NN O I-OtherScientificTerm
, NN O O
gradient NN O I-OtherScientificTerm
, NN O O
and NN O O
spectrum NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
, NN O O
and NN O O
use NN O O
feature NN O I-Method
parameter NN O I-Method
training NN O I-Method
to NN O O
robustly NN O O
classify NN O O
blurred NN O I-Material
images NN O I-Material
. NN O O

Our NN O O
blur NN O I-Method
detection NN O I-Method
is NN O O
based NN O O
on NN O O
image NN O I-OtherScientificTerm
patches NN O I-OtherScientificTerm
, NN O O
making NN O O
region-wise NN O I-Method
training NN O I-Method
and NN O I-Method
classification NN O I-Method
in NN O O
one NN O O
image NN O O
efficient NN O O
. NN O O

Extensive NN O O
experiments NN O O
show NN O O
that NN O O
our NN O O
method NN O I-Generic
works NN O O
satisfactorily NN O O
on NN O O
challenging NN O O
image NN O I-Material
data NN O I-Material
, NN O O
which NN O O
establishes NN O O
a NN O O
technical NN O O
foundation NN O O
for NN O O
solving NN O O
several NN O O
computer NN O I-Task
vision NN O I-Task
problems NN O I-Task
, NN O O
such NN O O
as NN O O
motion NN O I-Task
analysis NN O I-Task
and NN O O
image NN O I-Task
restoration NN O I-Task
, NN O O
using NN O O
the NN O O
blur NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

It NN O O
is NN O O
well-known NN O O
that NN O O
there NN O O
are NN O O
polysemous NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
like NN O O
sentence NN O O
whose NN O O
meaning NN O O
or NN O O
sense NN O O
depends NN O O
on NN O O
the NN O O
context NN O O
of NN O O
use NN O O
. NN O O

We NN O O
have NN O O
recently NN O O
reported NN O O
on NN O O
two NN O O
new NN O O
word-sense NN O I-Method
disambiguation NN O I-Method
systems NN O I-Method
, NN O O
one NN O I-Generic
trained NN O O
on NN O O
bilingual NN O I-Material
material NN O I-Material
-LRB- NN O O
the NN O O
Canadian NN O I-Material
Hansards NN O I-Material
-RRB- NN O O
and NN O O
the NN O O
other NN O I-Generic
trained NN O O
on NN O O
monolingual NN O I-Material
material NN O I-Material
-LRB- NN O O
Roget NN O I-Material
's NN O I-Material
Thesaurus NN O I-Material
and NN O O
Grolier NN O I-Material
's NN O I-Material
Encyclopedia NN O I-Material
-RRB- NN O O
. NN O O

As NN O O
this NN O O
work NN O O
was NN O O
nearing NN O O
completion NN O O
, NN O O
we NN O O
observed NN O O
a NN O O
very NN O O
strong NN O O
discourse NN O I-OtherScientificTerm
effect NN O O
. NN O O

That NN O O
is NN O O
, NN O O
if NN O O
a NN O O
polysemous NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
such NN O O
as NN O O
sentence NN O O
appears NN O O
two NN O O
or NN O O
more NN O O
times NN O O
in NN O O
a NN O O
well-written NN O I-OtherScientificTerm
discourse NN O I-OtherScientificTerm
, NN O O
it NN O O
is NN O O
extremely NN O O
likely NN O O
that NN O O
they NN O O
will NN O O
all NN O O
share NN O O
the NN O O
same NN O O
sense NN O O
. NN O O

This NN O O
paper NN O O
describes NN O O
an NN O O
experiment NN O O
which NN O O
confirmed NN O O
this NN O O
hypothesis NN O O
and NN O O
found NN O O
that NN O O
the NN O O
tendency NN O O
to NN O O
share NN O O
sense NN O O
in NN O O
the NN O O
same NN O O
discourse NN O I-OtherScientificTerm
is NN O O
extremely NN O O
strong NN O O
-LRB- NN O O
98 NN O O
% NN O O
-RRB- NN O O
. NN O O

This NN O O
result NN O O
can NN O O
be NN O O
used NN O O
as NN O O
an NN O O
additional NN O O
source NN O O
of NN O O
constraint NN O I-OtherScientificTerm
for NN O O
improving NN O O
the NN O O
performance NN O O
of NN O O
the NN O O
word-sense NN O I-Method
disambiguation NN O I-Method
algorithm NN O I-Method
. NN O O

In NN O O
addition NN O O
, NN O O
it NN O I-Generic
could NN O O
also NN O O
be NN O O
used NN O O
to NN O O
help NN O O
evaluate NN O O
disambiguation NN O I-Method
algorithms NN O I-Method
that NN O O
did NN O O
not NN O O
make NN O O
use NN O O
of NN O O
the NN O O
discourse NN O I-OtherScientificTerm
constraint NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

We NN O O
study NN O O
and NN O O
compare NN O O
two NN O O
novel NN O O
embedding NN O I-Method
methods NN O I-Method
for NN O O
segmenting NN O I-Task
feature NN O I-Task
points NN O I-Task
of NN O I-Task
piece-wise NN O I-Task
planar NN O I-Task
structures NN O I-Task
from NN O O
two NN O O
-LRB- NN O O
uncalibrated NN O O
-RRB- NN O O
perspective NN O O
images NN O O
. NN O O

We NN O O
show NN O O
that NN O O
a NN O O
set NN O O
of NN O O
different NN O O
homographies NN O I-OtherScientificTerm
can NN O O
be NN O O
embedded NN O O
in NN O O
different NN O O
ways NN O O
to NN O O
a NN O O
higher-dimensional NN O I-OtherScientificTerm
real NN O I-OtherScientificTerm
or NN O I-OtherScientificTerm
complex NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
, NN O O
so NN O O
that NN O O
each NN O O
homography NN O I-OtherScientificTerm
corresponds NN O O
to NN O O
either NN O O
a NN O O
complex NN O I-OtherScientificTerm
bilinear NN O I-OtherScientificTerm
form NN O I-OtherScientificTerm
or NN O O
a NN O O
real NN O I-OtherScientificTerm
quadratic NN O I-OtherScientificTerm
form NN O I-OtherScientificTerm
. NN O O

Each NN O O
embedding NN O O
reveals NN O O
different NN O O
algebraic NN O O
properties NN O O
and NN O O
relations NN O O
of NN O O
homo-graphies NN O I-OtherScientificTerm
. NN O O

We NN O O
give NN O O
a NN O O
closed-form NN O I-Method
segmentation NN O I-Method
solution NN O I-Method
for NN O O
each NN O O
case NN O O
by NN O O
utilizing NN O O
these NN O O
properties NN O O
based NN O O
on NN O O
subspace-segmentation NN O I-Method
methods NN O I-Method
. NN O O

These NN O O
theoretical NN O O
results NN O O
show NN O O
that NN O O
one NN O O
can NN O O
intrinsically NN O O
segment NN O O
a NN O O
piece-wise NN O I-OtherScientificTerm
planar NN O I-OtherScientificTerm
scene NN O I-OtherScientificTerm
from NN O O
2-D NN O I-Material
images NN O I-Material
without NN O O
explicitly NN O O
performing NN O O
any NN O O
3-D NN O I-Method
reconstruction NN O I-Method
. NN O O

The NN O O
resulting NN O O
segmentation NN O O
may NN O O
make NN O O
subsequent NN O O
3-D NN O I-Method
reconstruction NN O I-Method
much NN O O
better-conditioned NN O O
. NN O O

We NN O O
demonstrate NN O O
the NN O O
proposed NN O O
methods NN O O
with NN O O
some NN O O
convincing NN O O
experimental NN O O
results NN O O
. NN O O

-DOCSTART- NN O O

Background NN O I-Task
maintenance NN O I-Task
is NN O O
a NN O O
frequent NN O O
element NN O O
of NN O O
video NN O I-Task
surveillance NN O I-Task
systems NN O I-Task
. NN O O

We NN O O
develop NN O O
Wallflower NN O I-Method
, NN O O
a NN O O
three-component NN O I-Generic
system NN O I-Generic
for NN O O
background NN O I-Task
maintenance NN O I-Task
: NN O O
the NN O O
pixel-level NN O I-Method
component NN O I-Method
performs NN O O
Wiener NN O I-Method
filtering NN O I-Method
to NN O O
make NN O O
probabilistic NN O I-Task
predictions NN O I-Task
of NN O I-Task
the NN O I-Task
expected NN O I-Task
background NN O I-Task
; NN O O
the NN O O
region-level NN O I-Method
component NN O I-Method
fills NN O O
in NN O O
homogeneous NN O I-OtherScientificTerm
regions NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
foreground NN O I-OtherScientificTerm
objects NN O I-OtherScientificTerm
; NN O O
and NN O O
the NN O O
frame-level NN O I-Method
component NN O I-Method
detects NN O O
sudden NN O O
, NN O O
global NN O O
changes NN O O
in NN O O
the NN O O
image NN O O
and NN O O
swaps NN O O
in NN O O
better NN O O
approximations NN O O
of NN O O
the NN O O
background NN O O
. NN O O

We NN O O
compare NN O O
our NN O O
system NN O I-Generic
with NN O O
8 NN O O
other NN O O
background NN O I-Method
subtraction NN O I-Method
algorithms NN O I-Method
. NN O O

Wallflower NN O I-Method
is NN O O
shown NN O O
to NN O O
outperform NN O O
previous NN O O
algorithms NN O I-Generic
by NN O O
handling NN O O
a NN O O
greater NN O O
set NN O O
of NN O O
the NN O O
difficult NN O O
situations NN O O
that NN O O
can NN O O
occur NN O O
. NN O O

Finally NN O O
, NN O O
we NN O O
analyze NN O O
the NN O O
experimental NN O O
results NN O O
and NN O O
propose NN O O
normative NN O I-Method
principles NN O I-Method
for NN O O
background NN O I-Task
maintenance NN O I-Task
. NN O O

-DOCSTART- NN O O

Is NN O O
it NN O O
possible NN O O
to NN O O
use NN O O
out-of-domain NN O I-Material
acoustic NN O I-Material
training NN O I-Material
data NN O I-Material
to NN O O
improve NN O O
a NN O O
speech NN O I-Method
recognizer NN O I-Method
's NN O O
performance NN O O
on NN O O
a NN O O
speciic NN O O
, NN O O
independent NN O O
application NN O O
? NN O O

In NN O O
our NN O O
experiments NN O O
, NN O O
we NN O O
use NN O O
Wallstreet NN O I-Material
Journal NN O I-Material
-LRB- NN O I-Material
WSJ NN O I-Material
-RRB- NN O I-Material
data NN O I-Material
to NN O O
train NN O O
a NN O O
recognizer NN O I-Method
, NN O O
which NN O O
is NN O O
adapted NN O O
and NN O O
evaluated NN O O
in NN O O
the NN O O
Phonebook NN O I-Material
domain NN O I-Material
. NN O O

Apart NN O O
from NN O O
their NN O O
common NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
US NN O I-OtherScientificTerm
English NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
, NN O O
the NN O O
two NN O O
corpora NN O O
diier NN O O
in NN O O
many NN O O
important NN O O
respects NN O O
: NN O O
microphone NN O I-OtherScientificTerm
vs. NN O I-OtherScientificTerm
telephone NN O I-OtherScientificTerm
channel NN O I-OtherScientificTerm
, NN O O
continuous NN O I-OtherScientificTerm
speech NN O I-OtherScientificTerm
vs. NN O O
isolated NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
, NN O O
mismatch NN O O
i NN O O
n NN O O
s NN O O
p NN O O
e NN O O
a NN O O
k NN O O
i NN O O
n NN O O
g NN O O
r NN O O
a NN O O
t NN O O
e NN O O
. NN O O

This NN O O
paper NN O O
deals NN O O
with NN O O
two NN O O
questions NN O O
. NN O O

First NN O O
, NN O O
starting NN O O
from NN O O
the NN O O
WSJ-trained NN O I-Method
recognizer NN O I-Method
, NN O O
how NN O O
much NN O O
adaptation NN O I-Material
data NN O I-Material
-LRB- NN O O
taken NN O O
from NN O O
the NN O O
Phonebook NN O I-Material
training NN O I-Material
corpus NN O I-Material
-RRB- NN O O
is NN O O
necessary NN O O
to NN O O
achieve NN O O
a NN O O
reasonable NN O O
recognition NN O I-Task
performance NN O O
in NN O O
spite NN O O
of NN O O
the NN O O
high NN O O
degree NN O O
of NN O O
mismatch NN O I-OtherScientificTerm
? NN O O

Second NN O O
, NN O O
is NN O O
it NN O O
possible NN O O
to NN O O
improve NN O O
the NN O O
recognition NN O I-Task
performance NN O O
of NN O O
a NN O O
Phonebook-trained NN O I-Method
baseline NN O I-Method
acoustic NN O I-Method
model NN O I-Method
by NN O O
using NN O O
additional NN O O
out-of-domain NN O I-Material
training NN O I-Material
data NN O I-Material
? NN O O

The NN O O
paper NN O O
describes NN O O
the NN O O
adaptation NN O I-Method
and NN O I-Method
normalization NN O I-Method
techniques NN O I-Method
used NN O O
to NN O O
bridge NN O O
the NN O O
mismatch NN O O
b NN O O
e-tween NN O O
the NN O O
two NN O O
corpora NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
an NN O O
approach NN O I-Generic
to NN O O
full NN O I-Task
parsing NN O I-Task
suitable NN O O
for NN O O
Information NN O I-Task
Extraction NN O I-Task
from NN O O
texts NN O O
. NN O O

Sequences NN O O
of NN O O
cascades NN O O
of NN O O
rules NN O I-OtherScientificTerm
deterministically NN O O
analyze NN O O
the NN O O
text NN O O
, NN O O
building NN O O
unambiguous NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
. NN O O

Initially NN O O
basic NN O O
chunks NN O O
are NN O O
analyzed NN O O
; NN O O
then NN O O
argumental NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
are NN O O
recognized NN O O
; NN O O
finally NN O O
modifier NN O I-OtherScientificTerm
attachment NN O I-OtherScientificTerm
is NN O O
performed NN O O
and NN O O
the NN O O
global NN O I-OtherScientificTerm
parse NN O I-OtherScientificTerm
tree NN O I-OtherScientificTerm
is NN O O
built NN O O
. NN O O

The NN O O
approach NN O I-Generic
was NN O O
proven NN O O
to NN O O
work NN O O
for NN O O
three NN O O
languages NN O O
and NN O O
different NN O O
domains NN O O
. NN O O

It NN O I-Generic
was NN O O
implemented NN O O
in NN O O
the NN O O
IE NN O I-Method
module NN O I-Method
of NN O O
FACILE NN O I-Method
, NN O I-Method
a NN O I-Method
EU NN O I-Method
project NN O I-Method
for NN O I-Method
multilingual NN O I-Method
text NN O I-Method
classification NN O I-Method
and NN O I-Method
IE NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
corpus NN O I-Task
study NN O I-Task
that NN O O
explores NN O O
the NN O O
extent NN O O
to NN O O
which NN O O
captions NN O O
contribute NN O O
to NN O O
recognizing NN O O
the NN O O
intended NN O O
message NN O O
of NN O O
an NN O O
information NN O I-OtherScientificTerm
graphic NN O I-OtherScientificTerm
. NN O O

It NN O O
then NN O O
presents NN O O
an NN O O
implemented NN O O
graphic NN O I-Method
interpretation NN O I-Method
system NN O I-Method
that NN O O
takes NN O O
into NN O O
account NN O O
a NN O O
variety NN O O
of NN O O
communicative NN O I-Material
signals NN O I-Material
, NN O O
and NN O O
an NN O O
evaluation NN O O
study NN O O
showing NN O O
that NN O O
evidence NN O O
obtained NN O O
from NN O O
shallow NN O I-Method
processing NN O I-Method
of NN O O
the NN O O
graphic NN O I-Material
's NN O I-Material
caption NN O I-Material
has NN O O
a NN O O
significant NN O O
impact NN O O
on NN O O
the NN O O
system NN O I-Generic
's NN O O
success NN O O
. NN O O

This NN O O
work NN O O
is NN O O
part NN O O
of NN O O
a NN O O
larger NN O O
project NN O O
whose NN O O
goal NN O O
is NN O O
to NN O O
provide NN O O
sight-impaired NN O I-OtherScientificTerm
users NN O I-OtherScientificTerm
with NN O O
effective NN O O
access NN O O
to NN O O
information NN O I-OtherScientificTerm
graphics NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Graphical NN O I-Method
models NN O I-Method
such NN O O
as NN O O
Bayesian NN O I-Method
Networks NN O I-Method
-LRB- NN O I-Method
BNs NN O I-Method
-RRB- NN O I-Method
are NN O O
being NN O O
increasingly NN O O
applied NN O O
to NN O O
various NN O O
computer NN O I-Task
vision NN O I-Task
problems NN O I-Task
. NN O O

One NN O O
bottleneck NN O O
in NN O O
using NN O O
BN NN O I-Method
is NN O O
that NN O O
learning NN O O
the NN O O
BN NN O I-OtherScientificTerm
model NN O I-OtherScientificTerm
parameters NN O I-OtherScientificTerm
often NN O O
requires NN O O
a NN O O
large NN O O
amount NN O O
of NN O O
reliable NN O O
and NN O O
representative NN O I-Material
training NN O I-Material
data NN O I-Material
, NN O O
which NN O O
proves NN O O
to NN O O
be NN O O
difficult NN O O
to NN O O
acquire NN O O
for NN O O
many NN O O
computer NN O I-Task
vision NN O I-Task
tasks NN O I-Task
. NN O O

On NN O O
the NN O O
other NN O O
hand NN O O
, NN O O
there NN O O
is NN O O
often NN O O
available NN O O
qualitative NN O I-OtherScientificTerm
prior NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
about NN O O
the NN O O
model NN O I-Generic
. NN O O

Such NN O O
knowledge NN O I-Generic
comes NN O O
either NN O O
from NN O O
domain NN O I-OtherScientificTerm
experts NN O I-OtherScientificTerm
based NN O O
on NN O O
their NN O O
experience NN O O
or NN O O
from NN O O
various NN O O
physical NN O I-OtherScientificTerm
or NN O I-OtherScientificTerm
geometric NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
that NN O O
govern NN O O
the NN O O
objects NN O O
we NN O O
try NN O O
to NN O O
model NN O O
. NN O O

Unlike NN O O
the NN O O
quantitative NN O I-OtherScientificTerm
prior NN O I-OtherScientificTerm
, NN O O
the NN O O
qualitative NN O I-OtherScientificTerm
prior NN O I-OtherScientificTerm
is NN O O
often NN O O
ignored NN O O
due NN O O
to NN O O
the NN O O
difficulty NN O O
of NN O O
incorporating NN O O
them NN O I-Generic
into NN O O
the NN O O
model NN O I-Task
learning NN O I-Task
process NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
introduce NN O O
a NN O O
closed-form NN O I-Method
solution NN O I-Method
to NN O O
systematically NN O O
combine NN O O
the NN O O
limited NN O I-Material
training NN O I-Material
data NN O I-Material
with NN O O
some NN O O
generic NN O O
qualitative NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
for NN O O
BN NN O I-Method
parameter NN O I-Method
learning NN O I-Method
. NN O O

To NN O O
validate NN O O
our NN O O
method NN O I-Generic
, NN O O
we NN O O
compare NN O O
it NN O I-Generic
with NN O O
the NN O O
Maximum NN O I-Method
Likelihood NN O I-Method
-LRB- NN O I-Method
ML NN O I-Method
-RRB- NN O I-Method
estimation NN O I-Method
method NN O I-Method
under NN O O
sparse NN O I-Material
data NN O I-Material
and NN O O
with NN O O
the NN O O
Expectation NN O I-Method
Maximization NN O I-Method
-LRB- NN O I-Method
EM NN O I-Method
-RRB- NN O I-Method
algorithm NN O I-Method
under NN O O
incomplete NN O I-Material
data NN O I-Material
respectively NN O O
. NN O O

To NN O O
further NN O O
demonstrate NN O O
its NN O O
applications NN O O
for NN O O
computer NN O I-Task
vision NN O I-Task
, NN O O
we NN O O
apply NN O O
it NN O I-Generic
to NN O O
learn NN O O
a NN O O
BN NN O I-Method
model NN O I-Method
for NN O O
facial NN O I-Task
Action NN O I-Task
Unit NN O I-Task
-LRB- NN O I-Task
AU NN O I-Task
-RRB- NN O I-Task
recognition NN O I-Task
from NN O O
real NN O I-Material
image NN O I-Material
data NN O I-Material
. NN O O

The NN O O
experimental NN O O
results NN O O
show NN O O
that NN O O
with NN O O
simple NN O O
and NN O O
generic NN O I-OtherScientificTerm
qualitative NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
and NN O O
using NN O O
only NN O O
a NN O O
small NN O O
amount NN O O
of NN O O
training NN O I-Material
data NN O I-Material
, NN O O
our NN O O
method NN O I-Generic
can NN O O
robustly NN O O
and NN O O
accurately NN O O
estimate NN O O
the NN O O
BN NN O I-OtherScientificTerm
model NN O I-OtherScientificTerm
parameters NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
introduce NN O O
a NN O O
modal NN O I-Method
language NN O I-Method
LT NN O I-Method
for NN O O
imposing NN O O
constraints NN O I-OtherScientificTerm
on NN O I-OtherScientificTerm
trees NN O I-OtherScientificTerm
, NN O O
and NN O O
an NN O O
extension NN O I-Method
LT NN O I-Method
-LRB- NN O I-Method
LF NN O I-Method
-RRB- NN O I-Method
for NN O O
imposing NN O O
constraints NN O I-OtherScientificTerm
on NN O I-OtherScientificTerm
trees NN O I-OtherScientificTerm
decorated NN O I-OtherScientificTerm
with NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
. NN O O

The NN O O
motivation NN O O
for NN O O
introducing NN O O
these NN O O
languages NN O I-Generic
is NN O O
to NN O O
provide NN O O
tools NN O O
for NN O O
formalising NN O O
grammatical NN O I-Method
frameworks NN O I-Method
perspicuously NN O O
, NN O O
and NN O O
the NN O O
paper NN O O
illustrates NN O O
this NN O O
by NN O O
showing NN O O
how NN O O
the NN O O
leading NN O O
ideas NN O O
of NN O O
GPSG NN O I-Method
can NN O O
be NN O O
captured NN O O
in NN O O
LT NN O I-Method
-LRB- NN O I-Method
LF NN O I-Method
-RRB- NN O I-Method
. NN O O

In NN O O
addition NN O O
, NN O O
the NN O O
role NN O O
of NN O O
modal NN O I-Method
languages NN O I-Method
-LRB- NN O O
and NN O O
in NN O O
particular NN O O
, NN O O
what NN O O
we NN O O
have NN O O
called NN O O
as NN O O
constraint NN O I-Method
formalisms NN O I-Method
for NN O O
linguistic NN O O
theorising NN O O
is NN O O
discussed NN O O
in NN O O
some NN O O
detail NN O O
. NN O O

-DOCSTART- NN O O

Previous NN O O
research NN O O
has NN O O
demonstrated NN O O
the NN O O
utility NN O O
of NN O O
clustering NN O I-Method
in NN O O
inducing NN O I-Task
semantic NN O I-Task
verb NN O I-Task
classes NN O I-Task
from NN O O
undisambiguated NN O I-Material
corpus NN O I-Material
data NN O I-Material
. NN O O

We NN O O
describe NN O O
a NN O O
new NN O O
approach NN O I-Generic
which NN O O
involves NN O O
clustering NN O I-Task
subcategorization NN O I-Task
frame NN O I-Task
-LRB- NN O I-Task
SCF NN O I-Task
-RRB- NN O I-Task
distributions NN O I-Task
using NN O O
the NN O O
Information NN O I-Method
Bottleneck NN O I-Method
and NN O I-Method
nearest NN O I-Method
neighbour NN O I-Method
methods NN O I-Method
. NN O O

In NN O O
contrast NN O O
to NN O O
previous NN O O
work NN O O
, NN O O
we NN O O
particularly NN O O
focus NN O O
on NN O O
clustering NN O I-Task
polysemic NN O I-OtherScientificTerm
verbs NN O I-OtherScientificTerm
. NN O O

A NN O O
novel NN O O
evaluation NN O I-Generic
scheme NN O I-Generic
is NN O O
proposed NN O O
which NN O O
accounts NN O O
for NN O O
the NN O O
effect NN O O
of NN O O
polysemy NN O I-OtherScientificTerm
on NN O O
the NN O O
clusters NN O I-OtherScientificTerm
, NN O O
offering NN O O
us NN O O
a NN O O
good NN O O
insight NN O O
into NN O O
the NN O O
potential NN O O
and NN O O
limitations NN O O
of NN O O
semantically NN O I-Task
classifying NN O I-Task
undisambiguated NN O I-Task
SCF NN O I-Task
data NN O I-Task
. NN O O

-DOCSTART- NN O O

Due NN O O
to NN O O
the NN O O
capacity NN O O
of NN O O
pan-tilt-zoom NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
PTZ NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
cameras NN O I-OtherScientificTerm
to NN O O
simultaneously NN O O
cover NN O O
a NN O O
panoramic NN O I-OtherScientificTerm
area NN O I-OtherScientificTerm
and NN O O
maintain NN O O
high NN O I-OtherScientificTerm
resolution NN O I-OtherScientificTerm
imagery NN O I-OtherScientificTerm
, NN O O
researches NN O O
in NN O O
automated NN O I-Task
surveillance NN O I-Task
systems NN O I-Task
with NN O O
multiple NN O O
PTZ NN O I-OtherScientificTerm
cameras NN O I-OtherScientificTerm
have NN O O
become NN O O
increasingly NN O O
important NN O O
. NN O O

Most NN O O
existing NN O O
algorithms NN O I-Generic
require NN O O
the NN O O
prior NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
intrinsic NN O I-OtherScientificTerm
parameters NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
PTZ NN O I-OtherScientificTerm
camera NN O I-OtherScientificTerm
to NN O O
infer NN O O
the NN O O
relative NN O I-OtherScientificTerm
positioning NN O I-OtherScientificTerm
and NN O O
orientation NN O I-OtherScientificTerm
among NN O O
multiple NN O O
PTZ NN O I-OtherScientificTerm
cameras NN O I-OtherScientificTerm
. NN O O

To NN O O
overcome NN O O
this NN O O
limitation NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
novel NN O O
mapping NN O I-Method
algorithm NN O I-Method
that NN O O
derives NN O O
the NN O O
relative NN O I-OtherScientificTerm
positioning NN O I-OtherScientificTerm
and NN O O
orientation NN O I-OtherScientificTerm
between NN O O
two NN O O
PTZ NN O I-OtherScientificTerm
cameras NN O I-OtherScientificTerm
based NN O O
on NN O O
a NN O O
unified NN O I-Method
polynomial NN O I-Method
model NN O I-Method
. NN O O

This NN O O
reduces NN O O
the NN O O
dependence NN O O
on NN O O
the NN O O
knowledge NN O O
of NN O O
intrinsic NN O O
parameters NN O O
of NN O O
PTZ NN O I-OtherScientificTerm
camera NN O I-OtherScientificTerm
and NN O O
relative NN O I-OtherScientificTerm
positions NN O I-OtherScientificTerm
. NN O O

Experimental NN O O
results NN O O
demonstrate NN O O
that NN O O
our NN O O
proposed NN O O
algorithm NN O I-Generic
presents NN O O
substantially NN O O
reduced NN O O
computational NN O I-Metric
complexity NN O I-Metric
and NN O O
improved NN O O
flexibility NN O I-Metric
at NN O O
the NN O O
cost NN O O
of NN O O
slightly NN O O
decreased NN O O
pixel NN O I-Metric
accuracy NN O I-Metric
, NN O O
as NN O O
compared NN O O
with NN O O
the NN O O
work NN O O
of NN O O
Chen NN O O
and NN O O
Wang NN O O
. NN O O

This NN O O
slightly NN O O
decreased NN O O
pixel NN O I-Metric
accuracy NN O I-Metric
can NN O O
be NN O O
compensated NN O O
by NN O O
consistent NN O I-Method
labeling NN O I-Method
approaches NN O I-Method
without NN O O
added NN O O
cost NN O O
for NN O O
the NN O O
application NN O O
of NN O O
automated NN O I-Task
surveillance NN O I-Task
systems NN O I-Task
along NN O O
with NN O O
changing NN O O
configurations NN O O
and NN O O
a NN O O
larger NN O O
number NN O O
of NN O O
PTZ NN O I-OtherScientificTerm
cameras NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
new NN O O
two-pass NN O I-Generic
algorithm NN O I-Generic
for NN O O
Extra NN O I-Task
Large NN O I-Task
-LRB- NN O I-Task
more NN O I-Task
than NN O I-Task
1M NN O I-Task
words NN O I-Task
-RRB- NN O I-Task
Vocabulary NN O I-Task
COntinuous NN O I-Task
Speech NN O I-Task
recognition NN O I-Task
based NN O O
on NN O O
the NN O O
Information NN O I-Task
Retrieval NN O I-Task
-LRB- NN O I-Task
ELVIRCOS NN O I-Task
-RRB- NN O I-Task
. NN O O

The NN O O
principle NN O O
of NN O O
this NN O O
approach NN O I-Generic
is NN O O
to NN O O
decompose NN O O
a NN O O
recognition NN O I-Method
process NN O I-Method
into NN O O
two NN O O
passes NN O I-Generic
where NN O O
the NN O O
first NN O I-Generic
pass NN O I-Generic
builds NN O O
the NN O O
words NN O I-OtherScientificTerm
subset NN O I-OtherScientificTerm
for NN O O
the NN O O
second NN O I-Generic
pass NN O I-Generic
recognition NN O I-Generic
by NN O O
using NN O O
information NN O I-Method
retrieval NN O I-Method
procedure NN O I-Method
. NN O O

Word NN O I-Method
graph NN O I-Method
composition NN O I-Method
for NN O O
continuous NN O I-Material
speech NN O I-Material
is NN O O
presented NN O O
. NN O O

With NN O O
this NN O O
approach NN O I-Generic
a NN O O
high NN O O
performances NN O O
for NN O O
large NN O I-Task
vocabulary NN O I-Task
speech NN O I-Task
recognition NN O I-Task
can NN O O
be NN O O
obtained NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
our NN O O
work NN O O
on NN O O
classification NN O I-Task
of NN O I-Task
outdoor NN O I-Task
scenes NN O I-Task
. NN O O

First NN O O
, NN O O
images NN O O
are NN O O
partitioned NN O O
into NN O O
regions NN O O
using NN O O
one-class NN O I-Method
classification NN O I-Method
and NN O O
patch-based NN O I-Method
clustering NN O I-Method
algorithms NN O I-Method
where NN O O
one-class NN O I-Method
classifiers NN O I-Method
model NN O O
the NN O O
regions NN O O
with NN O O
relatively NN O O
uniform NN O I-OtherScientificTerm
color NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
texture NN O I-OtherScientificTerm
properties NN O I-OtherScientificTerm
, NN O O
and NN O O
clustering NN O I-Method
of NN O I-Method
patches NN O I-Method
aims NN O O
to NN O O
detect NN O O
structures NN O O
in NN O O
the NN O O
remaining NN O O
regions NN O O
. NN O O

Next NN O O
, NN O O
the NN O O
resulting NN O O
regions NN O O
are NN O O
clustered NN O O
to NN O O
obtain NN O O
a NN O O
codebook NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
region NN O I-OtherScientificTerm
types NN O I-OtherScientificTerm
, NN O O
and NN O O
two NN O O
models NN O I-Generic
are NN O O
constructed NN O O
for NN O O
scene NN O I-Method
representation NN O I-Method
: NN O O
a NN O O
'' NN O O
bag NN O O
of NN O O
individual NN O O
regions NN O O
'' NN O O
representation NN O O
where NN O O
each NN O O
region NN O O
is NN O O
regarded NN O O
separately NN O O
, NN O O
and NN O O
a NN O O
'' NN O O
bag NN O O
of NN O O
region NN O O
pairs NN O O
'' NN O O
representation NN O O
where NN O O
regions NN O O
with NN O O
particular NN O O
spatial NN O I-OtherScientificTerm
relationships NN O I-OtherScientificTerm
are NN O O
considered NN O O
together NN O O
. NN O O

Given NN O O
these NN O O
representations NN O O
, NN O O
scene NN O I-Task
classification NN O I-Task
is NN O O
done NN O O
using NN O O
Bayesian NN O I-Method
classifiers NN O I-Method
. NN O O

We NN O O
also NN O O
propose NN O O
a NN O O
novel NN O O
region NN O I-Method
selection NN O I-Method
algorithm NN O I-Method
that NN O O
identifies NN O O
region NN O O
types NN O O
that NN O O
are NN O O
frequently NN O O
found NN O O
in NN O O
a NN O O
particular NN O O
class NN O O
of NN O O
scenes NN O O
but NN O O
rarely NN O O
exist NN O O
in NN O O
other NN O O
classes NN O O
, NN O O
and NN O O
also NN O O
consistently NN O O
occur NN O O
together NN O O
in NN O O
the NN O O
same NN O O
class NN O O
of NN O O
scenes NN O O
. NN O O

Experiments NN O O
on NN O O
the NN O O
LabelMe NN O I-Material
data NN O I-Material
set NN O I-Material
showed NN O O
that NN O O
the NN O O
proposed NN O O
models NN O I-Generic
significantly NN O O
out-perform NN O O
a NN O O
baseline NN O I-Method
global NN O I-Method
feature-based NN O I-Method
approach NN O I-Method
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
introduce NN O O
a NN O O
generative NN O I-Method
probabilistic NN O I-Method
optical NN O I-Method
character NN O I-Method
recognition NN O I-Method
-LRB- NN O I-Method
OCR NN O I-Method
-RRB- NN O I-Method
model NN O I-Method
that NN O O
describes NN O O
an NN O O
end-to-end NN O O
process NN O O
in NN O O
the NN O O
noisy NN O I-Method
channel NN O I-Method
framework NN O I-Method
, NN O O
progressing NN O O
from NN O O
generation NN O O
of NN O O
true NN O O
text NN O O
through NN O O
its NN O O
transformation NN O O
into NN O O
the NN O O
noisy NN O O
output NN O O
of NN O O
an NN O O
OCR NN O I-OtherScientificTerm
system NN O I-OtherScientificTerm
. NN O O

The NN O O
model NN O I-Generic
is NN O O
designed NN O O
for NN O O
use NN O O
in NN O O
error NN O I-Task
correction NN O I-Task
, NN O O
with NN O O
a NN O O
focus NN O O
on NN O O
post-processing NN O I-Task
the NN O O
output NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
black-box NN O I-OtherScientificTerm
OCR NN O I-OtherScientificTerm
systems NN O I-OtherScientificTerm
in NN O O
order NN O O
to NN O O
make NN O O
it NN O I-Generic
more NN O O
useful NN O O
for NN O O
NLP NN O I-Task
tasks NN O I-Task
. NN O O

We NN O O
present NN O O
an NN O O
implementation NN O O
of NN O O
the NN O O
model NN O I-Generic
based NN O O
on NN O O
finite-state NN O I-Method
models NN O I-Method
, NN O O
demonstrate NN O O
the NN O O
model NN O I-Generic
's NN O O
ability NN O O
to NN O O
significantly NN O O
reduce NN O O
character NN O I-Metric
and NN O I-Metric
word NN O I-Metric
error NN O I-Metric
rate NN O I-Metric
, NN O O
and NN O O
provide NN O O
evaluation NN O O
results NN O O
involving NN O O
automatic NN O I-Task
extraction NN O I-Task
of NN O I-Task
translation NN O I-Task
lexicons NN O I-Task
from NN O O
printed NN O I-Material
text NN O I-Material
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
a NN O O
framework NN O I-Generic
for NN O O
word NN O I-Task
alignment NN O I-Task
based NN O O
on NN O O
log-linear NN O I-Method
models NN O I-Method
. NN O O

All NN O O
knowledge NN O I-Material
sources NN O I-Material
are NN O O
treated NN O O
as NN O O
feature NN O I-OtherScientificTerm
functions NN O I-OtherScientificTerm
, NN O O
which NN O O
depend NN O O
on NN O O
the NN O O
source NN O O
langauge NN O O
sentence NN O O
, NN O O
the NN O O
target NN O O
language NN O O
sentence NN O O
and NN O O
possible NN O O
additional NN O O
variables NN O O
. NN O O

Log-linear NN O I-Method
models NN O I-Method
allow NN O O
statistical NN O I-Method
alignment NN O I-Method
models NN O I-Method
to NN O O
be NN O O
easily NN O O
extended NN O O
by NN O O
incorporating NN O O
syntactic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
use NN O O
IBM NN O I-OtherScientificTerm
Model NN O I-OtherScientificTerm
3 NN O I-OtherScientificTerm
alignment NN O I-OtherScientificTerm
probabilities NN O I-OtherScientificTerm
, NN O O
POS NN O I-OtherScientificTerm
correspondence NN O I-OtherScientificTerm
, NN O O
and NN O O
bilingual NN O I-OtherScientificTerm
dictionary NN O I-OtherScientificTerm
coverage NN O I-OtherScientificTerm
as NN O O
features NN O I-OtherScientificTerm
. NN O O

Our NN O O
experiments NN O O
show NN O O
that NN O O
log-linear NN O I-Method
models NN O I-Method
significantly NN O O
outperform NN O O
IBM NN O I-Method
translation NN O I-Method
models NN O I-Method
. NN O O

-DOCSTART- NN O O

Hough NN O I-Method
voting NN O I-Method
in NN O O
a NN O O
geometric NN O I-OtherScientificTerm
transformation NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
allows NN O O
us NN O O
to NN O O
realize NN O O
spatial NN O I-Task
verification NN O I-Task
, NN O O
but NN O O
remains NN O O
sensitive NN O O
to NN O O
feature NN O I-OtherScientificTerm
detection NN O I-OtherScientificTerm
errors NN O I-OtherScientificTerm
because NN O O
of NN O O
the NN O O
inflexible NN O I-OtherScientificTerm
quan-tization NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
single NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
correspondences NN O I-OtherScientificTerm
. NN O O

To NN O O
handle NN O O
this NN O O
problem NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
new NN O O
method NN O I-Generic
, NN O O
called NN O O
adaptive NN O I-Method
dither NN O I-Method
voting NN O I-Method
, NN O O
for NN O O
robust NN O I-Task
spatial NN O I-Task
verification NN O I-Task
. NN O O

For NN O O
each NN O O
correspondence NN O O
, NN O O
instead NN O O
of NN O O
hard-mapping NN O O
it NN O O
to NN O O
a NN O O
single NN O O
transformation NN O O
, NN O O
the NN O O
method NN O I-Generic
augments NN O O
its NN O O
description NN O O
by NN O O
using NN O O
multiple NN O I-OtherScientificTerm
dithered NN O I-OtherScientificTerm
transformations NN O I-OtherScientificTerm
that NN O O
are NN O O
deterministically NN O O
generated NN O O
by NN O O
the NN O O
other NN O O
correspondences NN O O
. NN O O

The NN O O
method NN O I-Generic
reduces NN O O
the NN O O
probability NN O O
of NN O O
losing NN O O
correspondences NN O O
during NN O O
transformation NN O I-OtherScientificTerm
quantization NN O I-OtherScientificTerm
, NN O O
and NN O O
provides NN O O
high NN O O
robustness NN O O
as NN O O
regards NN O I-OtherScientificTerm
mismatches NN O I-OtherScientificTerm
by NN O O
imposing NN O O
three NN O O
geometric NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
on NN O O
the NN O O
dithering NN O I-OtherScientificTerm
process NN O I-OtherScientificTerm
. NN O O

We NN O O
also NN O O
propose NN O O
exploiting NN O O
the NN O O
non-uniformity NN O I-OtherScientificTerm
of NN O O
a NN O O
Hough NN O I-Method
histogram NN O I-Method
as NN O O
the NN O O
spatial NN O I-OtherScientificTerm
similarity NN O I-OtherScientificTerm
to NN O O
handle NN O O
multiple NN O I-OtherScientificTerm
matching NN O I-OtherScientificTerm
surfaces NN O I-OtherScientificTerm
. NN O O

Extensive NN O O
experiments NN O O
conducted NN O O
on NN O O
four NN O O
datasets NN O O
show NN O O
the NN O O
superiority NN O O
of NN O O
our NN O O
method NN O I-Generic
. NN O O

The NN O O
method NN O I-Generic
outperforms NN O O
its NN O O
state-of-the-art NN O O
counterparts NN O I-Generic
in NN O O
both NN O O
accuracy NN O I-Metric
and NN O O
scalability NN O I-Metric
, NN O O
especially NN O O
when NN O O
it NN O O
comes NN O O
to NN O O
the NN O O
retrieval NN O I-Task
of NN O I-Task
small NN O I-Task
, NN O I-Task
rotated NN O I-Task
objects NN O I-Task
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
a NN O O
novel NN O O
technique NN O I-Generic
called NN O O
bispectral NN O I-Method
photo-metric NN O I-Method
stereo NN O I-Method
that NN O O
makes NN O O
effective NN O O
use NN O O
of NN O O
fluorescence NN O I-OtherScientificTerm
for NN O O
shape NN O I-Task
reconstruction NN O I-Task
. NN O O

Fluorescence NN O I-OtherScientificTerm
is NN O O
a NN O O
common NN O O
phenomenon NN O O
occurring NN O O
in NN O O
many NN O O
objects NN O O
from NN O O
natural NN O I-OtherScientificTerm
gems NN O I-OtherScientificTerm
and NN O O
corals NN O O
, NN O O
to NN O O
fluorescent NN O I-OtherScientificTerm
dyes NN O I-OtherScientificTerm
used NN O O
in NN O O
clothing NN O O
. NN O O

One NN O O
of NN O O
the NN O O
important NN O O
characteristics NN O O
of NN O O
fluorescence NN O I-OtherScientificTerm
is NN O O
its NN O O
wavelength-shifting NN O O
behavior NN O O
: NN O O
fluorescent NN O I-OtherScientificTerm
materials NN O I-OtherScientificTerm
absorb NN O O
light NN O O
at NN O O
a NN O O
certain NN O O
wavelength NN O O
and NN O O
then NN O O
reemit NN O O
it NN O O
at NN O O
longer NN O O
wavelengths NN O O
. NN O O

Due NN O O
to NN O O
the NN O O
complexity NN O I-Metric
of NN O O
its NN O O
emission NN O I-Method
process NN O I-Method
, NN O O
fluo-rescence NN O I-OtherScientificTerm
tends NN O O
to NN O O
be NN O O
excluded NN O O
from NN O O
most NN O O
algorithms NN O I-Generic
in NN O O
computer NN O I-Task
vision NN O I-Task
and NN O O
image NN O I-Task
processing NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
show NN O O
that NN O O
there NN O O
is NN O O
a NN O O
strong NN O O
similarity NN O I-OtherScientificTerm
between NN O O
fluorescence NN O I-OtherScientificTerm
and NN O O
ideal NN O O
diffuse NN O I-OtherScientificTerm
reflection NN O I-OtherScientificTerm
and NN O O
that NN O O
fluorescence NN O I-OtherScientificTerm
can NN O O
provide NN O O
distinct NN O O
clues NN O O
on NN O O
how NN O O
to NN O O
estimate NN O O
an NN O O
object NN O O
's NN O O
shape NN O O
. NN O O

Moreover NN O O
, NN O O
fluorescence NN O I-OtherScientificTerm
's NN O I-OtherScientificTerm
wavelength-shifting NN O I-OtherScientificTerm
property NN O I-OtherScientificTerm
enables NN O O
us NN O O
to NN O O
estimate NN O O
the NN O O
shape NN O I-OtherScientificTerm
of NN O O
an NN O O
object NN O O
by NN O O
applying NN O O
photomet-ric NN O I-Method
stereo NN O I-Method
to NN O O
emission-only NN O I-Material
images NN O I-Material
without NN O O
suffering NN O O
from NN O O
specular NN O I-OtherScientificTerm
reflection NN O I-OtherScientificTerm
. NN O O

This NN O O
is NN O O
the NN O O
significant NN O O
advantage NN O O
of NN O O
the NN O O
fluorescence-based NN O I-Method
method NN O I-Method
over NN O O
previous NN O O
methods NN O I-Generic
based NN O O
on NN O O
reflection NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
an NN O O
approach NN O I-Generic
for NN O O
learning NN O O
a NN O O
visual NN O I-Method
representation NN O I-Method
from NN O O
the NN O O
raw NN O I-Material
spatiotemporal NN O I-Material
signals NN O I-Material
in NN O I-Material
videos NN O I-Material
. NN O O

Our NN O O
representation NN O I-Generic
is NN O O
learned NN O O
without NN O O
supervision NN O I-OtherScientificTerm
from NN O I-OtherScientificTerm
semantic NN O I-OtherScientificTerm
labels NN O I-OtherScientificTerm
. NN O O

We NN O O
formulate NN O O
our NN O O
method NN O I-Generic
as NN O O
an NN O O
unsupervised NN O I-Task
sequential NN O I-Task
verification NN O I-Task
task NN O I-Task
, NN O O
i.e. NN O O
, NN O O
we NN O O
determine NN O O
whether NN O O
a NN O O
sequence NN O O
of NN O O
frames NN O O
from NN O O
a NN O O
video NN O O
is NN O O
in NN O O
the NN O O
correct NN O O
temporal NN O I-OtherScientificTerm
order NN O I-OtherScientificTerm
. NN O O

With NN O O
this NN O O
simple NN O O
task NN O I-Generic
and NN O O
no NN O O
semantic NN O I-OtherScientificTerm
labels NN O I-OtherScientificTerm
, NN O O
we NN O O
learn NN O O
a NN O O
powerful NN O O
visual NN O I-Method
representation NN O I-Method
using NN O O
a NN O O
Convolutional NN O I-Method
Neural NN O I-Method
Network NN O I-Method
-LRB- NN O I-Method
CNN NN O I-Method
-RRB- NN O I-Method
. NN O O

The NN O O
representation NN O I-Generic
contains NN O O
complementary NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
to NN O O
that NN O O
learned NN O O
from NN O O
supervised NN O I-Material
image NN O I-Material
datasets NN O I-Material
like NN O O
ImageNet NN O I-Material
. NN O O

Qualitative NN O O
results NN O O
show NN O O
that NN O O
our NN O O
method NN O I-Generic
captures NN O O
information NN O O
that NN O O
is NN O O
temporally NN O O
varying NN O O
, NN O O
such NN O O
as NN O O
human NN O I-OtherScientificTerm
pose NN O I-OtherScientificTerm
. NN O O

When NN O O
used NN O O
as NN O O
pre-training NN O I-Method
for NN O O
action NN O I-Task
recognition NN O I-Task
, NN O O
our NN O I-Method
method NN O I-Method
gives NN O O
significant NN O O
gains NN O O
over NN O O
learning NN O I-Method
without NN O I-Method
external NN O I-Method
data NN O I-Method
on NN O O
benchmark NN O I-Generic
datasets NN O I-Generic
like NN O O
UCF101 NN O I-Material
and NN O O
HMDB51 NN O I-Material
. NN O O

To NN O O
demonstrate NN O O
its NN O O
sensitivity NN O O
to NN O O
human NN O I-OtherScientificTerm
pose NN O I-OtherScientificTerm
, NN O O
we NN O O
show NN O O
results NN O O
for NN O O
pose NN O I-Task
estimation NN O I-Task
on NN O O
the NN O O
FLIC NN O I-Material
and NN O I-Material
MPII NN O I-Material
datasets NN O I-Material
that NN O O
are NN O O
competitive NN O O
, NN O O
or NN O O
better NN O O
than NN O O
approaches NN O I-Generic
using NN O O
significantly NN O O
more NN O O
supervision NN O I-OtherScientificTerm
. NN O O

Our NN O I-OtherScientificTerm
method NN O I-OtherScientificTerm
can NN O O
be NN O O
combined NN O O
with NN O O
supervised NN O I-Method
representations NN O I-Method
to NN O O
provide NN O O
an NN O O
additional NN O O
boost NN O O
in NN O O
accuracy NN O I-Metric
. NN O O

-DOCSTART- NN O O

`` NN O O
To NN O O
explain NN O O
complex NN O O
phenomena NN O O
, NN O O
an NN O O
explanation NN O I-Method
system NN O I-Method
must NN O O
be NN O O
able NN O O
to NN O O
select NN O O
information NN O O
from NN O O
a NN O O
formal NN O O
representation NN O O
of NN O O
domain NN O O
knowledge NN O O
, NN O O
organize NN O O
the NN O O
selected NN O O
information NN O O
into NN O O
multisentential NN O I-OtherScientificTerm
discourse NN O I-OtherScientificTerm
plans NN O I-OtherScientificTerm
, NN O O
and NN O O
realize NN O O
the NN O O
discourse NN O I-OtherScientificTerm
plans NN O I-OtherScientificTerm
in NN O O
text NN O O
. NN O O

Although NN O O
recent NN O O
years NN O O
have NN O O
witnessed NN O O
significant NN O O
progress NN O O
in NN O O
the NN O O
development NN O O
of NN O O
sophisticated NN O O
computational NN O O
mechanisms NN O O
for NN O O
explanation NN O O
, NN O O
empirical NN O O
results NN O O
have NN O O
been NN O O
limited NN O O
. NN O O

This NN O O
paper NN O O
reports NN O O
on NN O O
a NN O O
seven-year NN O O
effort NN O O
to NN O O
empirically NN O O
study NN O O
explanation NN O I-Task
generation NN O I-Task
from NN O O
semantically NN O I-Material
rich NN O I-Material
, NN O I-Material
large-scale NN O I-Material
knowledge NN O I-Material
bases NN O I-Material
. NN O O

In NN O O
particular NN O O
, NN O O
it NN O O
describes NN O O
a NN O O
robust NN O I-Method
explanation NN O I-Method
system NN O I-Method
that NN O O
constructs NN O O
multisentential NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
multi-paragraph NN O I-OtherScientificTerm
explanations NN O I-OtherScientificTerm
from NN O O
the NN O O
a NN O O
large-scale NN O I-Material
knowledge NN O I-Material
base NN O I-Material
in NN O O
the NN O O
domain NN O O
of NN O O
botanical NN O I-Material
anatomy NN O I-Material
, NN O O
physiology NN O I-Material
, NN O O
and NN O O
development NN O I-Material
. NN O O

We NN O O
introduce NN O O
the NN O O
evaluation NN O I-Generic
methodology NN O I-Generic
and NN O O
describe NN O O
how NN O O
performance NN O O
was NN O O
assessed NN O O
with NN O O
this NN O O
methodology NN O I-Generic
in NN O O
the NN O O
most NN O O
extensive NN O O
empirical NN O O
evaluation NN O O
conducted NN O O
on NN O O
an NN O O
explanation NN O I-Method
system NN O I-Method
. NN O O

In NN O O
this NN O O
evaluation NN O I-Generic
, NN O O
scored NN O O
within NN O O
'' NN O O
`` NN O O
half NN O O
a NN O O
grade NN O O
'' NN O O
'' NN O O
of NN O O
domain NN O O
experts NN O O
, NN O O
and NN O O
its NN O O
performance NN O O
exceeded NN O O
that NN O O
of NN O O
one NN O O
of NN O O
the NN O O
domain NN O O
experts NN O O
. NN O O
'' NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
an NN O O
operable NN O I-Generic
definition NN O I-Generic
of NN O O
focus NN O O
which NN O O
is NN O O
argued NN O O
to NN O O
be NN O O
of NN O O
a NN O O
cognito-pragmatic NN O I-OtherScientificTerm
nature NN O I-OtherScientificTerm
and NN O O
explore NN O O
how NN O O
it NN O I-Generic
is NN O O
determined NN O O
in NN O O
discourse NN O I-OtherScientificTerm
in NN O O
a NN O O
formalized NN O O
manner NN O O
. NN O O

For NN O O
this NN O O
purpose NN O O
, NN O O
a NN O I-Method
file NN O I-Method
card NN O I-Method
model NN O I-Method
of NN O I-Method
discourse NN O I-Method
model NN O I-Method
and NN O I-Method
knowledge NN O I-OtherScientificTerm
store NN O I-OtherScientificTerm
is NN O O
introduced NN O O
enabling NN O O
the NN O O
decomposition NN O O
and NN O O
formal NN O O
representation NN O O
of NN O O
its NN O O
determination NN O I-OtherScientificTerm
process NN O I-OtherScientificTerm
as NN O O
a NN O O
programmable NN O I-Method
algorithm NN O I-Method
-LRB- NN O O
FDA NN O I-Method
-RRB- NN O O
. NN O O

Interdisciplinary NN O O
evidence NN O O
from NN O O
social NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
cognitive NN O I-OtherScientificTerm
psychology NN O I-OtherScientificTerm
is NN O O
cited NN O O
and NN O O
the NN O O
prospect NN O O
of NN O O
the NN O O
integration NN O O
of NN O O
focus NN O O
via NN O O
FDA NN O I-OtherScientificTerm
as NN O O
a NN O O
discourse-level NN O I-OtherScientificTerm
construct NN O I-OtherScientificTerm
into NN O O
speech NN O I-Method
synthesis NN O I-Method
systems NN O I-Method
, NN O O
in NN O O
particular NN O O
, NN O O
concept-to-speech NN O I-Method
systems NN O I-Method
, NN O O
is NN O O
also NN O O
briefly NN O O
discussed NN O O
. NN O O

-DOCSTART- NN O O

Conditional NN O I-Method
Random NN O I-Method
Field NN O I-Method
models NN O I-Method
have NN O O
proved NN O O
effective NN O O
for NN O O
several NN O O
low-level NN O I-Task
computer NN O I-Task
vision NN O I-Task
problems NN O I-Task
. NN O O

Inference NN O I-Task
in NN O O
these NN O O
models NN O I-Generic
involves NN O O
solving NN O O
a NN O O
combinatorial NN O I-Task
optimization NN O I-Task
problem NN O I-Task
, NN O O
with NN O O
methods NN O I-Generic
such NN O O
as NN O O
graph NN O I-Method
cuts NN O I-Method
, NN O O
belief NN O I-Method
propagation NN O I-Method
. NN O O

Although NN O O
several NN O O
methods NN O O
have NN O O
been NN O O
proposed NN O O
to NN O O
learn NN O O
the NN O O
model NN O I-OtherScientificTerm
parameters NN O I-OtherScientificTerm
from NN O O
training NN O O
data NN O O
, NN O O
they NN O O
suffer NN O O
from NN O O
various NN O O
drawbacks NN O O
. NN O O

Learning NN O O
these NN O O
parameters NN O I-Generic
involves NN O O
computing NN O O
the NN O O
partition NN O I-OtherScientificTerm
function NN O I-OtherScientificTerm
, NN O O
which NN O O
is NN O O
intractable NN O O
. NN O O

To NN O O
overcome NN O O
this NN O O
, NN O O
state-of-the-art NN O O
structured NN O I-Method
learning NN O I-Method
methods NN O I-Method
frame NN O O
the NN O O
problem NN O I-Generic
as NN O O
one NN O O
of NN O O
large NN O I-Task
margin NN O I-Task
estimation NN O I-Task
. NN O O

Iterative NN O I-Method
solutions NN O I-Method
have NN O O
been NN O O
proposed NN O O
to NN O O
solve NN O O
the NN O O
resulting NN O O
convex NN O I-Task
optimization NN O I-Task
problem NN O I-Task
. NN O O

Each NN O O
iteration NN O O
involves NN O O
solving NN O O
an NN O O
inference NN O I-Task
problem NN O I-Task
over NN O O
all NN O O
the NN O O
labels NN O O
, NN O O
which NN O O
limits NN O O
the NN O O
efficiency NN O O
of NN O O
these NN O O
structured NN O I-Generic
methods NN O I-Generic
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
present NN O O
an NN O O
efficient NN O O
large NN O I-Method
margin NN O I-Method
piece-wise NN O I-Method
learning NN O I-Method
method NN O I-Method
which NN O O
is NN O O
widely NN O O
applicable NN O O
. NN O O

We NN O O
show NN O O
how NN O O
the NN O O
resulting NN O O
optimization NN O I-Task
problem NN O I-Task
can NN O O
be NN O O
reduced NN O O
to NN O O
an NN O O
equivalent NN O O
convex NN O I-Task
problem NN O I-Task
with NN O O
a NN O O
small NN O O
number NN O O
of NN O O
constraints NN O O
, NN O O
and NN O O
solve NN O O
it NN O O
using NN O O
an NN O O
efficient NN O O
scheme NN O O
. NN O O

Our NN O O
method NN O I-Generic
is NN O O
both NN O O
memory NN O O
and NN O O
computationally NN O O
efficient NN O O
. NN O O

We NN O O
show NN O O
results NN O O
on NN O O
publicly NN O O
available NN O O
standard NN O O
datasets NN O O
. NN O O

-DOCSTART- NN O O

Interpreting NN O I-Task
metaphors NN O I-Task
is NN O O
an NN O O
integral NN O O
and NN O O
inescapable NN O O
process NN O O
in NN O O
human NN O I-Task
understanding NN O I-Task
of NN O I-Task
natural NN O I-Task
language NN O I-Task
. NN O O

This NN O O
paper NN O O
discusses NN O O
a NN O O
method NN O I-Generic
of NN O O
analyzing NN O I-Task
metaphors NN O I-Task
based NN O O
on NN O O
the NN O O
existence NN O O
of NN O O
a NN O O
small NN O O
number NN O O
of NN O O
generalized NN O I-Method
metaphor NN O I-Method
mappings NN O I-Method
. NN O O

Each NN O O
generalized NN O I-OtherScientificTerm
metaphor NN O I-OtherScientificTerm
contains NN O O
a NN O O
recognition NN O I-Method
network NN O I-Method
, NN O O
a NN O O
basic NN O I-Method
mapping NN O I-Method
, NN O O
additional NN O O
transfer NN O I-Method
mappings NN O I-Method
, NN O O
and NN O O
an NN O O
implicit NN O I-Method
intention NN O I-Method
component NN O I-Method
. NN O O

It NN O O
is NN O O
argued NN O O
that NN O O
the NN O O
method NN O I-Generic
reduces NN O O
metaphor NN O I-Task
interpretation NN O I-Task
from NN O O
a NN O O
reconstruction NN O I-Task
to NN O O
a NN O O
recognition NN O I-Task
task NN O I-Task
. NN O O

Implications NN O O
towards NN O O
automating NN O O
certain NN O O
aspects NN O O
of NN O O
language NN O I-Task
learning NN O I-Task
are NN O O
also NN O O
discussed NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
study NN O O
presents NN O O
a NN O O
method NN O I-Method
to NN O I-Method
automatically NN O I-Method
acquire NN O I-Method
paraphrases NN O I-Method
using NN O O
bilingual NN O I-Material
corpora NN O I-Material
, NN O O
which NN O O
utilizes NN O O
the NN O O
bilingual NN O I-OtherScientificTerm
dependency NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
obtained NN O O
by NN O O
projecting NN O O
a NN O O
monolingual NN O I-OtherScientificTerm
dependency NN O I-OtherScientificTerm
parse NN O I-OtherScientificTerm
onto NN O O
the NN O O
other NN O O
language NN O O
sentence NN O O
based NN O O
on NN O O
statistical NN O I-Method
alignment NN O I-Method
techniques NN O I-Method
. NN O O

Since NN O O
the NN O O
paraphrasing NN O I-Method
method NN O I-Method
is NN O O
capable NN O O
of NN O O
clearly NN O O
disambiguating NN O O
the NN O O
sense NN O O
of NN O O
an NN O O
original NN O O
phrase NN O O
using NN O O
the NN O O
bilingual NN O I-OtherScientificTerm
context NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
dependency NN O I-OtherScientificTerm
relation NN O I-OtherScientificTerm
, NN O O
it NN O O
would NN O O
be NN O O
possible NN O O
to NN O O
obtain NN O O
interchangeable NN O O
paraphrases NN O I-OtherScientificTerm
under NN O O
a NN O O
given NN O O
context NN O O
. NN O O

Also NN O O
, NN O O
we NN O O
provide NN O O
an NN O O
advanced NN O O
method NN O I-Generic
to NN O O
acquire NN O O
generalized NN O I-OtherScientificTerm
translation NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
using NN O O
the NN O O
extracted NN O O
paraphrases NN O I-OtherScientificTerm
. NN O O

We NN O O
applied NN O O
the NN O O
method NN O I-Generic
to NN O O
acquire NN O O
the NN O O
generalized NN O I-OtherScientificTerm
translation NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
for NN O O
Korean-English NN O I-Task
translation NN O I-Task
. NN O O

Through NN O O
experiments NN O O
with NN O O
parallel NN O I-Material
corpora NN O I-Material
of NN O I-Material
a NN O I-Material
Korean NN O I-Material
and NN O I-Material
English NN O I-Material
language NN O I-Material
pairs NN O I-Material
, NN O O
we NN O O
show NN O O
that NN O O
our NN O O
paraphrasing NN O I-Method
method NN O I-Method
effectively NN O O
extracts NN O O
paraphrases NN O I-OtherScientificTerm
with NN O O
high NN O O
precision NN O I-Metric
, NN O O
94.3 NN O O
% NN O O
and NN O O
84.6 NN O O
% NN O O
respectively NN O O
for NN O O
Korean NN O I-Material
and NN O O
English NN O I-Material
, NN O O
and NN O O
the NN O O
translation NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
extracted NN O O
from NN O O
the NN O O
bilingual NN O I-Material
corpora NN O I-Material
could NN O O
be NN O O
generalized NN O O
successfully NN O O
using NN O O
the NN O O
paraphrases NN O I-OtherScientificTerm
with NN O O
the NN O O
12.5 NN O O
% NN O O
compression NN O I-Metric
ratio NN O I-Metric
. NN O O

-DOCSTART- NN O O

We NN O O
provide NN O O
a NN O O
logical NN O I-Method
definition NN O I-Method
of NN O I-Method
Minimalist NN O I-Method
grammars NN O I-Method
, NN O O
that NN O O
are NN O O
Stabler NN O I-Method
's NN O I-Method
formalization NN O I-Method
of NN O I-Method
Chomsky NN O I-Method
's NN O I-Method
minimalist NN O I-Method
program NN O I-Method
. NN O O

Our NN O O
logical NN O I-Method
definition NN O I-Method
leads NN O O
to NN O O
a NN O O
neat NN O O
relation NN O O
to NN O O
categorial NN O I-Method
grammar NN O I-Method
, NN O O
-LRB- NN O O
yielding NN O O
a NN O O
treatment NN O O
of NN O O
Montague NN O I-OtherScientificTerm
semantics NN O I-OtherScientificTerm
-RRB- NN O O
, NN O O
a NN O O
parsing-as-deduction NN O I-Method
in NN O O
a NN O O
resource NN O I-OtherScientificTerm
sensitive NN O I-OtherScientificTerm
logic NN O I-OtherScientificTerm
, NN O O
and NN O O
a NN O O
learning NN O I-Method
algorithm NN O I-Method
from NN O O
structured NN O I-Material
data NN O I-Material
-LRB- NN O O
based NN O O
on NN O O
a NN O O
typing-algorithm NN O I-Method
and NN O O
type-unification NN O I-Method
-RRB- NN O O
. NN O O

Here NN O O
we NN O O
emphasize NN O O
the NN O O
connection NN O O
to NN O O
Montague NN O I-OtherScientificTerm
semantics NN O I-OtherScientificTerm
which NN O O
can NN O O
be NN O O
viewed NN O O
as NN O O
a NN O O
formal NN O I-OtherScientificTerm
computation NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
logical NN O I-OtherScientificTerm
form NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

There NN O O
are NN O O
several NN O O
approaches NN O I-Generic
that NN O O
model NN O O
information NN O I-Task
extraction NN O I-Task
as NN O O
a NN O O
token NN O I-Task
classification NN O I-Task
task NN O I-Task
, NN O O
using NN O O
various NN O O
tagging NN O I-Method
strategies NN O I-Method
to NN O O
combine NN O O
multiple NN O O
tokens NN O O
. NN O O

We NN O O
describe NN O O
the NN O O
tagging NN O I-Method
strategies NN O I-Method
that NN O O
can NN O O
be NN O O
found NN O O
in NN O O
the NN O O
literature NN O O
and NN O O
evaluate NN O O
their NN O O
relative NN O O
performances NN O O
. NN O O

We NN O O
also NN O O
introduce NN O O
a NN O O
new NN O O
strategy NN O I-Generic
, NN O O
called NN O O
Begin/After NN O I-Method
tagging NN O I-Method
or NN O O
BIA NN O I-Method
, NN O O
and NN O O
show NN O O
that NN O O
it NN O I-Generic
is NN O O
competitive NN O O
to NN O O
the NN O O
best NN O O
other NN O O
strategies NN O I-Generic
. NN O O

-DOCSTART- NN O O

The NN O O
PRC NN O I-Method
Adaptive NN O I-Method
Knowledge-based NN O I-Method
Text NN O I-Method
Understanding NN O I-Method
System NN O I-Method
-LRB- NN O I-Method
PAKTUS NN O I-Method
-RRB- NN O I-Method
has NN O O
been NN O O
under NN O O
development NN O O
as NN O O
an NN O O
Independent NN O O
Research NN O O
and NN O O
Development NN O O
project NN O O
at NN O O
PRC NN O O
since NN O O
1984 NN O O
. NN O O

The NN O O
objective NN O O
is NN O O
a NN O O
generic NN O O
system NN O I-Generic
of NN O O
tools NN O O
, NN O O
including NN O O
a NN O O
core NN O I-OtherScientificTerm
English NN O I-OtherScientificTerm
lexicon NN O I-OtherScientificTerm
, NN O O
grammar NN O I-Method
, NN O O
and NN O O
concept NN O I-Method
representations NN O I-Method
, NN O O
for NN O O
building NN O O
natural NN O I-Method
language NN O I-Method
processing NN O I-Method
-LRB- NN O I-Method
NLP NN O I-Method
-RRB- NN O I-Method
systems NN O I-Method
for NN O O
text NN O I-Task
understanding NN O I-Task
. NN O O

Systems NN O O
built NN O O
with NN O O
PAKTUS NN O I-Method
are NN O O
intended NN O O
to NN O O
generate NN O O
input NN O O
to NN O O
knowledge NN O I-Task
based NN O I-Task
systems NN O I-Task
ordata NN O O
base NN O O
systems NN O O
. NN O O

Input NN O O
to NN O O
the NN O O
NLP NN O I-Method
system NN O I-Method
is NN O O
typically NN O O
derived NN O O
from NN O O
an NN O O
existing NN O O
electronic NN O I-Material
message NN O I-Material
stream NN O I-Material
, NN O O
such NN O O
as NN O O
a NN O O
news NN O I-Material
wire NN O I-Material
. NN O O

PAKTUS NN O I-Method
supports NN O O
the NN O O
adaptation NN O O
of NN O O
the NN O O
generic NN O O
core NN O O
to NN O O
a NN O O
variety NN O O
of NN O O
domains NN O O
: NN O O
JINTACCS NN O I-Material
messages NN O I-Material
, NN O O
RAINFORM NN O I-Material
messages NN O I-Material
, NN O O
news NN O I-OtherScientificTerm
reports NN O I-OtherScientificTerm
about NN O O
a NN O O
specific NN O O
type NN O O
of NN O O
event NN O I-OtherScientificTerm
, NN O O
such NN O O
as NN O O
financial NN O I-Material
transfers NN O I-Material
or NN O O
terrorist NN O I-Material
acts NN O I-Material
, NN O O
etc. NN O O
, NN O O
by NN O O
acquiring NN O O
sublanguage NN O I-Method
and NN O I-Method
domain-specific NN O I-Method
grammar NN O I-Method
, NN O O
words NN O I-Generic
, NN O O
conceptual NN O I-OtherScientificTerm
mappings NN O I-OtherScientificTerm
, NN O O
and NN O O
discourse NN O I-OtherScientificTerm
patterns NN O I-OtherScientificTerm
. NN O O

The NN O O
long-term NN O O
goal NN O O
is NN O O
a NN O O
system NN O O
that NN O O
can NN O O
support NN O O
the NN O O
processing NN O O
of NN O O
relatively NN O O
long NN O O
discourses NN O O
in NN O O
domains NN O O
that NN O O
are NN O O
fairly NN O O
broad NN O O
with NN O O
a NN O O
high NN O O
rate NN O O
of NN O O
success NN O O
. NN O O

-DOCSTART- NN O O

A NN O O
major NN O O
axis NN O O
of NN O O
research NN O O
at NN O O
LIMSI NN O O
is NN O O
directed NN O O
at NN O O
multilingual NN O I-Task
, NN O I-Task
speaker-independent NN O I-Task
, NN O I-Task
large NN O I-Task
vocabulary NN O I-Task
speech NN O I-Task
dictation NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
the NN O O
LIMSI NN O I-Method
recognizer NN O I-Method
which NN O O
was NN O O
evaluated NN O O
in NN O O
the NN O O
ARPA NN O I-Material
NOV93 NN O I-Material
CSR NN O I-Material
test NN O I-Material
is NN O O
described NN O O
, NN O O
and NN O O
experimental NN O O
results NN O O
on NN O O
the NN O O
WSJ NN O I-Material
and NN O I-Material
BREF NN O I-Material
corpora NN O I-Material
under NN O O
closely NN O O
matched NN O O
conditions NN O O
are NN O O
reported NN O O
. NN O O

For NN O O
both NN O O
corpora NN O I-Generic
word NN O I-Task
recognition NN O I-Task
experiments NN O O
were NN O O
carried NN O O
out NN O O
with NN O O
vocabularies NN O O
containing NN O O
up NN O O
to NN O O
20k NN O O
words NN O O
. NN O O

The NN O O
recognizer NN O I-Generic
makes NN O O
use NN O O
of NN O O
continuous NN O I-Method
density NN O I-Method
HMM NN O I-Method
with NN O O
Gaussian NN O I-Method
mixture NN O I-Method
for NN O O
acoustic NN O I-Task
modeling NN O I-Task
and NN O O
n-gram NN O I-Method
statistics NN O I-Method
estimated NN O O
on NN O O
the NN O O
newspaper NN O I-Material
texts NN O I-Material
for NN O O
language NN O I-Task
modeling NN O I-Task
. NN O O

The NN O O
recognizer NN O I-Generic
uses NN O O
a NN O O
time-synchronous NN O I-Method
graph-search NN O I-Method
strategy NN O I-Method
which NN O O
is NN O O
shown NN O O
to NN O O
still NN O O
be NN O O
viable NN O O
with NN O O
a NN O O
20k-word NN O O
vocabulary NN O O
when NN O O
used NN O O
with NN O O
bigram NN O I-Method
back-off NN O I-Method
language NN O I-Method
models NN O I-Method
. NN O O

A NN O O
second NN O O
forward NN O O
pass NN O O
, NN O O
which NN O O
makes NN O O
use NN O O
of NN O O
a NN O O
word NN O I-OtherScientificTerm
graph NN O I-OtherScientificTerm
generated NN O O
with NN O O
the NN O O
bigram NN O I-Method
, NN O O
incorporates NN O O
a NN O O
trigram NN O I-Method
language NN O I-Method
model NN O I-Method
. NN O O

Acoustic NN O I-Task
modeling NN O I-Task
uses NN O O
cepstrum-based NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
, NN O O
context-dependent NN O I-Method
phone NN O I-Method
models NN O I-Method
-LRB- NN O I-Method
intra NN O I-Method
and NN O I-Method
interword NN O I-Method
-RRB- NN O I-Method
, NN O O
phone NN O I-Method
duration NN O I-Method
models NN O I-Method
, NN O O
and NN O O
sex-dependent NN O I-Method
models NN O I-Method
. NN O O

-DOCSTART- NN O O

The NN O O
co-occurrence NN O I-OtherScientificTerm
pattern NN O I-OtherScientificTerm
, NN O O
a NN O O
combination NN O O
of NN O O
binary NN O I-OtherScientificTerm
or NN O I-OtherScientificTerm
local NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
, NN O O
is NN O O
more NN O O
discriminative NN O O
than NN O O
individual NN O O
features NN O O
and NN O O
has NN O O
shown NN O O
its NN O O
advantages NN O O
in NN O O
object NN O I-Task
, NN O I-Task
scene NN O I-Task
, NN O I-Task
and NN O I-Task
action NN O I-Task
recognition NN O I-Task
. NN O O

We NN O O
discuss NN O O
two NN O O
types NN O O
of NN O O
co-occurrence NN O I-OtherScientificTerm
patterns NN O I-OtherScientificTerm
that NN O O
are NN O O
complementary NN O O
to NN O O
each NN O O
other NN O O
, NN O O
the NN O O
conjunction NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
AND NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
disjunction NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
OR NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
binary NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
. NN O O

The NN O O
necessary NN O O
condition NN O O
of NN O O
identifying NN O I-Task
discriminative NN O I-OtherScientificTerm
co-occurrence NN O I-OtherScientificTerm
patterns NN O I-OtherScientificTerm
is NN O O
firstly NN O O
provided NN O O
. NN O O

Then NN O O
we NN O O
propose NN O O
a NN O O
novel NN O O
data NN O I-Method
mining NN O I-Method
method NN O I-Method
to NN O O
efficiently NN O O
discover NN O O
the NN O O
optimal NN O I-OtherScientificTerm
co-occurrence NN O I-OtherScientificTerm
pattern NN O I-OtherScientificTerm
with NN O O
minimum NN O I-OtherScientificTerm
empirical NN O I-OtherScientificTerm
error NN O I-OtherScientificTerm
, NN O O
despite NN O O
the NN O O
noisy NN O I-Material
training NN O I-Material
dataset NN O I-Material
. NN O O

This NN O O
mining NN O I-Generic
procedure NN O I-Generic
of NN O O
AND NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
OR NN O I-OtherScientificTerm
patterns NN O I-OtherScientificTerm
is NN O O
readily NN O O
integrated NN O O
to NN O O
boosting NN O I-Method
, NN O O
which NN O O
improves NN O O
the NN O O
generalization NN O I-Metric
ability NN O I-Metric
over NN O O
the NN O O
conventional NN O O
boosting NN O I-Method
decision NN O I-Method
trees NN O I-Method
and NN O O
boosting NN O I-Method
decision NN O I-Method
stumps NN O I-Method
. NN O O

Our NN O O
versatile NN O O
experiments NN O O
on NN O O
object NN O I-Task
, NN O I-Task
scene NN O I-Task
, NN O I-Task
and NN O I-Task
action NN O I-Task
cat-egorization NN O I-Task
validate NN O O
the NN O O
advantages NN O O
of NN O O
the NN O O
discovered NN O O
dis-criminative NN O I-OtherScientificTerm
co-occurrence NN O I-OtherScientificTerm
patterns NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Empirical NN O O
experience NN O O
and NN O O
observations NN O O
have NN O O
shown NN O O
us NN O O
when NN O O
powerful NN O O
and NN O O
highly NN O O
tunable NN O O
classifiers NN O I-Method
such NN O O
as NN O O
maximum NN O I-Method
entropy NN O I-Method
classifiers NN O I-Method
, NN O O
boosting NN O I-Method
and NN O O
SVMs NN O I-Method
are NN O O
applied NN O O
to NN O O
language NN O I-Task
processing NN O I-Task
tasks NN O I-Task
, NN O O
it NN O O
is NN O O
possible NN O O
to NN O O
achieve NN O O
high NN O O
accuracies NN O O
, NN O O
but NN O O
eventually NN O O
their NN O O
performances NN O O
all NN O O
tend NN O O
to NN O O
plateau NN O O
out NN O O
at NN O O
around NN O O
the NN O O
same NN O O
point NN O O
. NN O O

To NN O O
further NN O O
improve NN O O
performance NN O O
, NN O O
various NN O O
error NN O I-Method
correction NN O I-Method
mechanisms NN O I-Method
have NN O O
been NN O O
developed NN O O
, NN O O
but NN O O
in NN O O
practice NN O O
, NN O O
most NN O O
of NN O O
them NN O O
can NN O O
not NN O O
be NN O O
relied NN O O
on NN O O
to NN O O
predictably NN O O
improve NN O O
performance NN O O
on NN O O
unseen NN O O
data NN O O
; NN O O
indeed NN O O
, NN O O
depending NN O O
upon NN O O
the NN O O
test NN O O
set NN O O
, NN O O
they NN O O
are NN O O
as NN O O
likely NN O O
to NN O O
degrade NN O O
accuracy NN O O
as NN O O
to NN O O
improve NN O O
it NN O O
. NN O O

This NN O O
problem NN O O
is NN O O
especially NN O O
severe NN O O
if NN O O
the NN O O
base NN O I-Generic
classifier NN O I-Generic
has NN O O
already NN O O
been NN O O
finely NN O O
tuned NN O O
. NN O O

In NN O O
recent NN O O
work NN O O
, NN O O
we NN O O
introduced NN O O
N-fold NN O I-Method
Templated NN O I-Method
Piped NN O I-Method
Correction NN O I-Method
, NN O I-Method
or NN O I-Method
NTPC NN O I-Method
-LRB- NN O I-Method
`` NN O I-Method
nitpick NN O I-Method
'' NN O I-Method
-RRB- NN O I-Method
, NN O O
an NN O O
intriguing NN O O
error NN O I-Method
corrector NN O I-Method
that NN O O
is NN O O
designed NN O O
to NN O O
work NN O O
in NN O O
these NN O O
extreme NN O O
operating NN O O
conditions NN O O
. NN O O

Despite NN O O
its NN O O
simplicity NN O O
, NN O O
it NN O I-Generic
consistently NN O O
and NN O O
robustly NN O O
improves NN O O
the NN O O
accuracy NN O O
of NN O O
existing NN O O
highly NN O O
accurate NN O O
base NN O I-Generic
models NN O I-Generic
. NN O O

This NN O O
paper NN O O
investigates NN O O
some NN O O
of NN O O
the NN O O
more NN O O
surprising NN O O
claims NN O O
made NN O O
by NN O O
NTPC NN O I-Method
, NN O O
and NN O O
presents NN O O
experiments NN O O
supporting NN O O
an NN O O
Occam NN O I-OtherScientificTerm
's NN O I-OtherScientificTerm
Razor NN O I-OtherScientificTerm
argument NN O I-OtherScientificTerm
that NN O O
more NN O O
complex NN O O
models NN O O
are NN O O
damaging NN O O
or NN O O
unnecessary NN O O
in NN O O
practice NN O O
. NN O O

-DOCSTART- NN O O

A NN O O
flexible NN O I-Method
parser NN O I-Method
can NN O O
deal NN O O
with NN O O
input NN O O
that NN O O
deviates NN O O
from NN O O
its NN O O
grammar NN O O
, NN O O
in NN O O
addition NN O O
to NN O O
input NN O O
that NN O O
conforms NN O O
to NN O O
it NN O I-Generic
. NN O O

Ideally NN O O
, NN O O
such NN O O
a NN O O
parser NN O I-Method
will NN O O
correct NN O O
the NN O O
deviant NN O O
input NN O O
: NN O O
sometimes NN O O
, NN O O
it NN O O
will NN O O
be NN O O
unable NN O O
to NN O O
correct NN O O
it NN O O
at NN O O
all NN O O
; NN O O
at NN O O
other NN O O
times NN O O
, NN O O
correction NN O O
will NN O O
be NN O O
possible NN O O
, NN O O
but NN O O
only NN O O
to NN O O
within NN O O
a NN O O
range NN O O
of NN O O
ambiguous NN O O
possibilities NN O O
. NN O O

This NN O O
paper NN O O
is NN O O
concerned NN O O
with NN O O
such NN O O
ambiguous NN O O
situations NN O O
, NN O O
and NN O O
with NN O O
making NN O O
it NN O O
as NN O O
easy NN O O
as NN O O
possible NN O O
for NN O O
the NN O O
ambiguity NN O I-OtherScientificTerm
to NN O O
be NN O O
resolved NN O O
through NN O O
consultation NN O O
with NN O O
the NN O O
user NN O O
of NN O O
the NN O O
parser NN O I-Method
- NN O O
we NN O O
presume NN O O
interactive NN O O
use NN O O
. NN O O

We NN O O
show NN O O
the NN O O
importance NN O O
of NN O O
asking NN O O
the NN O O
user NN O O
for NN O O
clarification NN O O
in NN O O
as NN O O
focused NN O O
a NN O O
way NN O O
as NN O O
possible NN O O
. NN O O

Focused NN O O
interaction NN O O
of NN O O
this NN O O
kind NN O O
is NN O O
facilitated NN O O
by NN O O
a NN O O
construction-specific NN O I-Method
approach NN O I-Method
to NN O O
flexible NN O I-Task
parsing NN O I-Task
, NN O O
with NN O O
specialized NN O I-Method
parsing NN O I-Method
techniques NN O I-Method
for NN O O
each NN O O
type NN O O
of NN O O
construction NN O I-Task
, NN O O
and NN O O
specialized NN O O
ambiguity NN O I-Method
representations NN O I-Method
for NN O O
each NN O O
type NN O O
of NN O O
ambiguity NN O I-OtherScientificTerm
that NN O O
a NN O O
particular NN O O
construction NN O I-Task
can NN O O
give NN O O
rise NN O O
to NN O O
. NN O O

A NN O O
construction-specific NN O I-Method
approach NN O I-Method
also NN O O
aids NN O O
in NN O O
task-specific NN O I-Task
language NN O I-Task
development NN O I-Task
by NN O O
allowing NN O O
a NN O O
language NN O O
definition NN O O
that NN O O
is NN O O
natural NN O O
in NN O O
terms NN O O
of NN O O
the NN O O
task NN O O
domain NN O O
to NN O O
be NN O O
interpreted NN O O
directly NN O O
without NN O O
compilation NN O O
into NN O O
a NN O O
uniform NN O I-Method
grammar NN O I-Method
formalism NN O I-Method
, NN O O
thus NN O O
greatly NN O O
speeding NN O O
the NN O O
testing NN O O
of NN O O
changes NN O O
to NN O O
the NN O O
language NN O O
definition NN O O
. NN O O

-DOCSTART- NN O O

A NN O O
proposal NN O I-Generic
to NN O O
deal NN O O
with NN O O
French NN O I-OtherScientificTerm
tenses NN O I-OtherScientificTerm
in NN O O
the NN O O
framework NN O O
of NN O O
Discourse NN O I-Method
Representation NN O I-Method
Theory NN O I-Method
is NN O O
presented NN O O
, NN O O
as NN O O
it NN O I-Generic
has NN O O
been NN O O
implemented NN O O
for NN O O
a NN O O
fragment NN O O
at NN O O
the NN O O
IMS NN O I-Task
. NN O O

It NN O I-Generic
is NN O O
based NN O O
on NN O O
the NN O O
theory NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
tenses NN O I-OtherScientificTerm
of NN O O
H. NN O O
Kamp NN O O
and NN O O
Ch NN O O
. NN O O

Rohrer NN O O
. NN O O

Instead NN O O
of NN O O
using NN O O
operators NN O I-Generic
to NN O O
express NN O O
the NN O O
meaning NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
tenses NN O I-OtherScientificTerm
the NN O O
Reichenbachian NN O O
point NN O O
of NN O O
view NN O O
is NN O O
adopted NN O O
and NN O O
refined NN O O
such NN O O
that NN O O
the NN O O
impact NN O O
of NN O O
the NN O O
tenses NN O O
with NN O O
respect NN O O
to NN O O
the NN O O
meaning NN O O
of NN O O
the NN O O
text NN O O
is NN O O
understood NN O O
as NN O O
contribution NN O O
to NN O O
the NN O O
integration NN O O
of NN O O
the NN O O
events NN O O
of NN O O
a NN O O
sentence NN O O
in NN O O
the NN O O
event NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
of NN O O
the NN O O
preceeding NN O O
text NN O O
. NN O O

Thereby NN O O
a NN O O
system NN O I-Method
of NN O I-Method
relevant NN O I-Method
times NN O I-Method
provided NN O O
by NN O O
the NN O O
preceeding NN O I-OtherScientificTerm
text NN O I-OtherScientificTerm
and NN O O
by NN O O
the NN O O
temporal NN O I-OtherScientificTerm
adverbials NN O I-OtherScientificTerm
of NN O O
the NN O O
sentence NN O O
being NN O O
processed NN O O
is NN O O
used NN O O
. NN O O

This NN O O
system NN O I-Generic
consists NN O O
of NN O O
one NN O O
or NN O O
more NN O O
reference NN O I-OtherScientificTerm
times NN O I-OtherScientificTerm
and NN O O
temporal NN O I-OtherScientificTerm
perspective NN O I-OtherScientificTerm
times NN O I-OtherScientificTerm
, NN O O
the NN O O
speech NN O I-OtherScientificTerm
time NN O I-OtherScientificTerm
and NN O O
the NN O O
location NN O I-OtherScientificTerm
time NN O I-OtherScientificTerm
. NN O O

The NN O O
special NN O O
interest NN O O
of NN O O
our NN O O
proposal NN O O
is NN O O
to NN O O
establish NN O O
a NN O O
plausible NN O O
choice NN O O
of NN O O
anchors NN O O
for NN O O
the NN O O
new NN O O
event NN O O
out NN O O
of NN O O
the NN O O
system NN O I-Method
of NN O I-Method
relevant NN O I-Method
times NN O I-Method
and NN O O
to NN O O
update NN O O
this NN O O
system NN O I-Method
of NN O I-Method
temporal NN O I-Method
coordinates NN O I-Method
correctly NN O O
. NN O O

The NN O O
problem NN O O
of NN O O
choice NN O O
is NN O O
largely NN O O
neglected NN O O
in NN O O
the NN O O
literature NN O O
. NN O O

In NN O O
opposition NN O O
to NN O O
the NN O O
approach NN O O
of NN O O
Kamp NN O O
and NN O O
Rohrer NN O O
the NN O O
exact NN O O
meaning NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
tenses NN O I-OtherScientificTerm
is NN O O
fixed NN O O
by NN O O
the NN O O
resolution NN O I-Method
component NN O I-Method
and NN O O
not NN O O
in NN O O
the NN O O
process NN O O
of NN O O
syntactic NN O I-Method
analysis NN O I-Method
. NN O O

-DOCSTART- NN O O

The NN O O
work NN O O
presented NN O O
in NN O O
this NN O O
paper NN O O
is NN O O
the NN O O
first NN O O
step NN O O
in NN O O
a NN O O
project NN O O
which NN O O
aims NN O O
to NN O O
cluster NN O O
and NN O O
summarise NN O O
electronic NN O I-Task
discussions NN O I-Task
in NN O O
the NN O O
context NN O O
of NN O O
help-desk NN O I-Task
applications NN O I-Task
. NN O O

The NN O O
eventual NN O O
objective NN O O
of NN O O
this NN O O
project NN O O
is NN O O
to NN O O
use NN O O
these NN O O
summaries NN O O
to NN O O
assist NN O O
help-desk NN O O
users NN O O
and NN O O
operators NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
identify NN O O
features NN O I-OtherScientificTerm
of NN O O
electronic NN O I-Task
discussions NN O I-Task
that NN O O
influence NN O O
the NN O O
clustering NN O I-Method
process NN O I-Method
, NN O O
and NN O O
offer NN O O
a NN O O
filtering NN O I-Method
mechanism NN O I-Method
that NN O O
removes NN O O
undesirable NN O O
influences NN O O
. NN O O

We NN O O
tested NN O O
the NN O O
clustering NN O I-Method
and NN O I-Method
filtering NN O I-Method
processes NN O I-Method
on NN O O
electronic NN O I-Material
newsgroup NN O I-Material
discussions NN O I-Material
, NN O O
and NN O O
evaluated NN O O
their NN O O
performance NN O O
by NN O O
means NN O O
of NN O O
two NN O O
experiments NN O I-Generic
: NN O O
coarse-level NN O I-Task
clustering NN O I-Task
simple NN O O
information NN O I-Task
retrieval NN O I-Task
. NN O O

-DOCSTART- NN O O

The NN O O
paper NN O O
presents NN O O
a NN O O
method NN O I-Generic
for NN O O
word NN O I-Task
sense NN O I-Task
disambiguation NN O I-Task
based NN O O
on NN O O
parallel NN O I-Material
corpora NN O I-Material
. NN O O

The NN O O
method NN O I-Generic
exploits NN O O
recent NN O O
advances NN O O
in NN O O
word NN O I-Task
alignment NN O I-Task
and NN O O
word NN O I-Task
clustering NN O I-Task
based NN O O
on NN O O
automatic NN O I-Task
extraction NN O I-Task
of NN O I-Task
translation NN O I-Task
equivalents NN O I-Task
and NN O O
being NN O O
supported NN O O
by NN O O
available NN O O
aligned NN O I-Material
wordnets NN O I-Material
for NN O O
the NN O O
languages NN O O
in NN O O
the NN O O
corpus NN O O
. NN O O

The NN O O
wordnets NN O I-Material
are NN O O
aligned NN O O
to NN O O
the NN O O
Princeton NN O I-Material
Wordnet NN O I-Material
, NN O O
according NN O O
to NN O O
the NN O O
principles NN O O
established NN O O
by NN O O
EuroWordNet NN O I-Material
. NN O O

The NN O O
evaluation NN O O
of NN O O
the NN O O
WSD NN O I-Method
system NN O I-Method
, NN O O
implementing NN O O
the NN O O
method NN O I-Generic
described NN O O
herein NN O O
showed NN O O
very NN O O
encouraging NN O O
results NN O O
. NN O O

The NN O O
same NN O O
system NN O I-Generic
used NN O O
in NN O O
a NN O O
validation NN O O
mode NN O O
, NN O O
can NN O O
be NN O O
used NN O O
to NN O O
check NN O O
and NN O O
spot NN O O
alignment NN O I-OtherScientificTerm
errors NN O I-OtherScientificTerm
in NN O I-OtherScientificTerm
multilingually NN O I-Material
aligned NN O I-Material
wordnets NN O I-Material
as NN O O
BalkaNet NN O I-Material
and NN O O
EuroWordNet NN O I-Material
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
investigates NN O O
critical NN O O
configurations NN O O
for NN O O
projective NN O I-Task
reconstruction NN O I-Task
from NN O O
multiple NN O O
images NN O I-Material
taken NN O O
by NN O O
a NN O O
camera NN O O
moving NN O O
in NN O O
a NN O O
straight NN O O
line NN O O
. NN O O

Projective NN O I-Task
reconstruction NN O I-Task
refers NN O O
to NN O O
a NN O O
determination NN O O
of NN O O
the NN O O
3D NN O I-OtherScientificTerm
geometrical NN O I-OtherScientificTerm
configuration NN O I-OtherScientificTerm
of NN O O
a NN O O
set NN O O
of NN O O
3D NN O I-OtherScientificTerm
points NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
cameras NN O I-OtherScientificTerm
, NN O O
given NN O O
only NN O O
correspondences NN O O
between NN O O
points NN O O
in NN O O
the NN O O
images NN O O
. NN O O

A NN O O
configuration NN O O
of NN O O
points NN O O
and NN O O
cameras NN O O
is NN O O
critical NN O O
if NN O O
it NN O O
can NN O O
not NN O O
be NN O O
determined NN O O
uniquely NN O O
-LRB- NN O O
up NN O O
to NN O O
a NN O O
projective NN O O
transform NN O O
-RRB- NN O O
from NN O O
the NN O O
image NN O I-OtherScientificTerm
coordinates NN O I-OtherScientificTerm
of NN O O
the NN O O
points NN O O
. NN O O

It NN O O
is NN O O
shown NN O O
that NN O O
a NN O O
configuration NN O O
consisting NN O O
of NN O O
any NN O O
number NN O O
of NN O O
cameras NN O O
lying NN O O
on NN O O
a NN O O
straight NN O O
line NN O O
, NN O O
and NN O O
any NN O O
number NN O O
of NN O O
points NN O O
lying NN O O
on NN O O
a NN O O
twisted NN O O
cubic NN O O
constitutes NN O O
a NN O O
critical NN O O
configuration NN O O
. NN O O

An NN O O
alternative NN O O
configuration NN O O
consisting NN O O
of NN O O
a NN O O
set NN O O
of NN O O
points NN O O
and NN O O
cameras NN O O
all NN O O
lying NN O O
on NN O O
a NN O O
rational NN O I-OtherScientificTerm
quartic NN O I-OtherScientificTerm
curve NN O I-OtherScientificTerm
exists NN O O
. NN O O

-DOCSTART- NN O O

Porting NN O O
a NN O O
Natural NN O I-Task
Language NN O I-Task
Processing NN O I-Task
-LRB- NN O I-Task
NLP NN O I-Task
-RRB- NN O I-Task
system NN O I-Task
to NN O O
a NN O O
new NN O I-Material
domain NN O I-Material
remains NN O O
one NN O O
of NN O O
the NN O O
bottlenecks NN O O
in NN O O
syntactic NN O I-Task
parsing NN O I-Task
, NN O O
because NN O O
of NN O O
the NN O O
amount NN O O
of NN O O
effort NN O O
required NN O O
to NN O O
fix NN O O
gaps NN O O
in NN O O
the NN O O
lexicon NN O O
, NN O O
and NN O O
to NN O O
attune NN O O
the NN O O
existing NN O O
grammar NN O I-Method
to NN O O
the NN O O
idiosyncracies NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
new NN O I-OtherScientificTerm
sublanguage NN O I-OtherScientificTerm
. NN O O

This NN O O
paper NN O O
shows NN O O
how NN O O
the NN O O
process NN O O
of NN O O
fitting NN O O
a NN O O
lexicalized NN O I-Method
grammar NN O I-Method
to NN O O
a NN O O
domain NN O O
can NN O O
be NN O O
automated NN O O
to NN O O
a NN O O
great NN O O
extent NN O O
by NN O O
using NN O O
a NN O O
hybrid NN O I-Method
system NN O I-Method
that NN O O
combines NN O O
traditional NN O O
knowledge-based NN O I-Method
techniques NN O I-Method
with NN O O
a NN O O
corpus-based NN O I-Method
approach NN O I-Method
. NN O O

-DOCSTART- NN O O

Chart NN O O
parsing NN O O
is NN O O
directional NN O O
in NN O O
the NN O O
sense NN O O
that NN O O
it NN O O
works NN O O
from NN O O
the NN O O
starting NN O O
point NN O O
-LRB- NN O O
usually NN O O
the NN O O
beginning NN O O
of NN O O
the NN O O
sentence NN O O
-RRB- NN O O
extending NN O O
its NN O O
activity NN O O
usually NN O O
in NN O O
a NN O O
rightward NN O O
manner NN O O
. NN O O

We NN O O
shall NN O O
introduce NN O O
the NN O O
concept NN O O
of NN O O
a NN O O
chart NN O O
that NN O O
works NN O O
outward NN O O
from NN O O
islands NN O O
and NN O O
makes NN O O
sense NN O O
of NN O O
as NN O O
much NN O O
of NN O O
the NN O O
sentence NN O O
as NN O O
it NN O O
is NN O O
actually NN O O
possible NN O O
, NN O O
and NN O O
after NN O O
that NN O O
will NN O O
lead NN O O
to NN O O
predictions NN O O
of NN O O
missing NN O O
fragments NN O O
. NN O O

So NN O O
, NN O O
for NN O O
any NN O O
place NN O O
where NN O O
the NN O O
easily NN O O
identifiable NN O O
fragments NN O O
occur NN O O
in NN O O
the NN O O
sentence NN O O
, NN O O
the NN O O
process NN O O
will NN O O
extend NN O O
to NN O O
both NN O O
the NN O O
left NN O O
and NN O O
the NN O O
right NN O O
of NN O O
the NN O O
islands NN O O
, NN O O
until NN O O
possibly NN O O
completely NN O O
missing NN O O
fragments NN O O
are NN O O
reached NN O O
. NN O O

At NN O O
that NN O O
point NN O O
, NN O O
by NN O O
virtue NN O O
of NN O O
the NN O O
fact NN O O
that NN O O
both NN O O
a NN O O
left NN O O
and NN O O
a NN O O
right NN O O
context NN O O
were NN O O
found NN O O
, NN O O
heuristics NN O O
can NN O O
be NN O O
introduced NN O O
that NN O O
predict NN O O
the NN O O
nature NN O O
of NN O O
the NN O O
missing NN O O
fragments NN O O
. NN O O

-DOCSTART- NN O O

Unification NN O I-Method
is NN O O
often NN O O
the NN O O
appropriate NN O O
method NN O I-Generic
for NN O O
expressing NN O O
relations NN O I-Task
between NN O I-Task
representations NN O I-Task
in NN O O
the NN O O
form NN O O
of NN O O
feature NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
; NN O O
however NN O O
, NN O O
there NN O O
are NN O O
circumstances NN O O
in NN O O
which NN O O
a NN O O
different NN O O
approach NN O I-Generic
is NN O O
desirable NN O O
. NN O O

A NN O O
declarative NN O I-Method
formalism NN O I-Method
is NN O O
presented NN O O
which NN O O
permits NN O O
direct NN O I-OtherScientificTerm
mappings NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
one NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
into NN O I-OtherScientificTerm
another NN O I-OtherScientificTerm
, NN O O
and NN O O
illustrative NN O O
examples NN O O
are NN O O
given NN O O
of NN O O
its NN O O
application NN O O
to NN O O
areas NN O O
of NN O O
current NN O O
interest NN O O
. NN O O

-DOCSTART- NN O O

To NN O O
support NN O O
engaging NN O O
human NN O O
users NN O O
in NN O O
robust NN O O
, NN O O
mixed-initiative NN O I-Method
speech NN O I-Method
dialogue NN O I-Method
interactions NN O I-Method
which NN O O
reach NN O O
beyond NN O O
current NN O O
capabilities NN O O
in NN O O
dialogue NN O I-Method
systems NN O I-Method
, NN O O
the NN O O
DARPA NN O O
Communicator NN O O
program NN O O
-LSB- NN O O
1 NN O O
-RSB- NN O O
is NN O O
funding NN O O
the NN O O
development NN O O
of NN O O
a NN O O
distributed NN O I-Method
message-passing NN O I-Method
infrastructure NN O I-Method
for NN O O
dialogue NN O I-Method
systems NN O I-Method
which NN O O
all NN O O
Communicator NN O O
participants NN O O
are NN O O
using NN O O
. NN O O

In NN O O
this NN O O
presentation NN O O
, NN O O
we NN O O
describe NN O O
the NN O O
features NN O O
of NN O O
and NN O O
requirements NN O O
for NN O O
a NN O O
genuinely NN O O
useful NN O O
software NN O O
infrastructure NN O O
for NN O O
this NN O O
purpose NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
a NN O O
novel NN O O
limited-memory NN O I-Method
stochastic NN O I-Method
block NN O I-Method
BFGS NN O I-Method
update NN O I-Method
for NN O O
incorporating NN O I-Task
enriched NN O I-Task
curvature NN O I-Task
information NN O I-Task
in NN O I-Task
stochastic NN O I-Task
approximation NN O I-Task
methods NN O I-Task
. NN O O

In NN O O
our NN O O
method NN O I-Generic
, NN O O
the NN O O
estimate NN O O
of NN O O
the NN O O
inverse NN O I-OtherScientificTerm
Hessian NN O I-OtherScientificTerm
matrix NN O I-OtherScientificTerm
that NN O O
is NN O O
maintained NN O O
by NN O O
it NN O I-Generic
, NN O O
is NN O O
updated NN O O
at NN O O
each NN O O
iteration NN O O
using NN O O
a NN O O
sketch NN O O
of NN O O
the NN O O
Hessian NN O I-OtherScientificTerm
, NN O O
i.e. NN O O
, NN O O
a NN O O
randomly NN O I-OtherScientificTerm
generated NN O I-OtherScientificTerm
compressed NN O I-OtherScientificTerm
form NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
Hessian NN O I-OtherScientificTerm
. NN O O

We NN O O
propose NN O O
several NN O O
sketching NN O I-Method
strategies NN O I-Method
, NN O O
present NN O O
a NN O O
new NN O O
quasi-Newton NN O I-Method
method NN O I-Method
that NN O O
uses NN O O
stochastic NN O I-Method
block NN O I-Method
BFGS NN O I-Method
updates NN O I-Method
combined NN O O
with NN O O
the NN O O
variance NN O I-Method
reduction NN O I-Method
approach NN O I-Method
SVRG NN O I-Method
to NN O O
compute NN O O
batch NN O I-OtherScientificTerm
stochastic NN O I-OtherScientificTerm
gradients NN O I-OtherScientificTerm
, NN O O
and NN O O
prove NN O O
linear NN O I-OtherScientificTerm
convergence NN O I-OtherScientificTerm
of NN O O
the NN O O
resulting NN O O
method NN O I-Generic
. NN O O

Numerical NN O O
tests NN O O
on NN O O
large-scale NN O I-Task
logistic NN O I-Task
regression NN O I-Task
problems NN O I-Task
reveal NN O O
that NN O O
our NN O O
method NN O I-Generic
is NN O O
more NN O O
robust NN O O
and NN O O
substantially NN O O
outperforms NN O O
current NN O O
state-of-the-art NN O I-Generic
methods NN O I-Generic
. NN O O

-DOCSTART- NN O O

The NN O O
goal NN O O
of NN O O
this NN O O
research NN O O
is NN O O
to NN O O
develop NN O O
a NN O O
spoken NN O I-Method
language NN O I-Method
system NN O I-Method
that NN O O
will NN O O
demonstrate NN O O
the NN O O
usefulness NN O O
of NN O O
voice NN O I-Material
input NN O I-Material
for NN O O
interactive NN O I-Task
problem NN O I-Task
solving NN O I-Task
. NN O O

The NN O O
system NN O O
will NN O O
accept NN O O
continuous NN O I-Material
speech NN O I-Material
, NN O O
and NN O O
will NN O O
handle NN O O
multiple NN O O
speakers NN O O
without NN O O
explicit NN O O
speaker NN O O
enrollment NN O O
. NN O O

Combining NN O O
speech NN O I-Task
recognition NN O I-Task
and NN O O
natural NN O I-Task
language NN O I-Task
processing NN O I-Task
to NN O O
achieve NN O O
speech NN O I-Task
understanding NN O I-Task
, NN O O
the NN O O
system NN O I-Generic
will NN O O
be NN O O
demonstrated NN O O
in NN O O
an NN O O
application NN O O
domain NN O O
relevant NN O O
to NN O O
the NN O O
DoD NN O O
. NN O O

The NN O O
objective NN O O
of NN O O
this NN O O
project NN O O
is NN O O
to NN O O
develop NN O O
a NN O O
robust NN O I-Method
and NN O I-Method
high-performance NN O I-Method
speech NN O I-Method
recognition NN O I-Method
system NN O I-Method
using NN O O
a NN O O
segment-based NN O I-Method
approach NN O I-Method
to NN O O
phonetic NN O I-Task
recognition NN O I-Task
. NN O O

The NN O O
recognition NN O I-Method
system NN O I-Method
will NN O O
eventually NN O O
be NN O O
integrated NN O O
with NN O O
natural NN O I-Task
language NN O I-Task
processing NN O I-Task
to NN O O
achieve NN O O
spoken NN O I-Task
language NN O I-Task
understanding NN O I-Task
. NN O O

-DOCSTART- NN O O

Spelling-checkers NN O I-Task
have NN O O
become NN O O
an NN O O
integral NN O O
part NN O O
of NN O O
most NN O O
text NN O I-Task
processing NN O I-Task
software NN O I-Task
. NN O O

From NN O O
different NN O O
reasons NN O O
among NN O O
which NN O O
the NN O O
speed NN O O
of NN O O
processing NN O O
prevails NN O O
they NN O I-Generic
are NN O O
usually NN O O
based NN O O
on NN O O
dictionaries NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
forms NN O I-OtherScientificTerm
instead NN O O
of NN O O
words NN O O
. NN O O

This NN O O
approach NN O I-Generic
is NN O O
sufficient NN O O
for NN O O
languages NN O I-Material
with NN O O
little NN O O
inflection NN O I-OtherScientificTerm
such NN O O
as NN O O
English NN O I-Material
, NN O O
but NN O O
fails NN O O
for NN O O
highly NN O I-Material
inflective NN O I-Material
languages NN O I-Material
such NN O O
as NN O O
Czech NN O I-Material
, NN O O
Russian NN O I-Material
, NN O O
Slovak NN O I-Material
or NN O O
other NN O O
Slavonic NN O I-Material
languages NN O I-Material
. NN O O

We NN O O
have NN O O
developed NN O O
a NN O O
special NN O O
method NN O I-Generic
for NN O O
describing NN O O
inflection NN O I-OtherScientificTerm
for NN O O
the NN O O
purpose NN O O
of NN O O
building NN O O
spelling-checkers NN O I-Task
for NN O O
such NN O O
languages NN O I-Generic
. NN O O

The NN O O
speed NN O O
of NN O O
the NN O O
resulting NN O O
program NN O I-Generic
lies NN O O
somewhere NN O O
in NN O O
the NN O O
middle NN O O
of NN O O
the NN O O
scale NN O O
of NN O O
existing NN O O
spelling-checkers NN O I-Method
for NN O O
English NN O I-Material
and NN O O
the NN O O
main NN O O
dictionary NN O O
fits NN O O
into NN O O
the NN O O
standard NN O O
360K NN O O
floppy NN O O
, NN O O
whereas NN O O
the NN O O
number NN O O
of NN O O
recognized NN O O
word NN O O
forms NN O O
exceeds NN O O
6 NN O O
million NN O O
-LRB- NN O O
for NN O O
Czech NN O I-Material
-RRB- NN O O
. NN O O

Further NN O O
, NN O O
a NN O O
special NN O O
method NN O I-Generic
has NN O O
been NN O O
developed NN O O
for NN O O
easy NN O O
word NN O I-Task
classification NN O I-Task
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
a NN O O
new NN O O
HMM NN O I-Method
tagger NN O I-Method
that NN O O
exploits NN O O
context NN O O
on NN O O
both NN O O
sides NN O O
of NN O O
a NN O O
word NN O O
to NN O O
be NN O O
tagged NN O O
, NN O O
and NN O O
evaluate NN O O
it NN O I-Generic
in NN O O
both NN O O
the NN O O
unsupervised NN O I-Task
and NN O I-Task
supervised NN O I-Task
case NN O I-Task
. NN O O

Along NN O O
the NN O O
way NN O O
, NN O O
we NN O O
present NN O O
the NN O O
first NN O O
comprehensive NN O O
comparison NN O O
of NN O O
unsupervised NN O I-Method
methods NN O I-Method
for NN O O
part-of-speech NN O I-Task
tagging NN O I-Task
, NN O O
noting NN O O
that NN O O
published NN O O
results NN O O
to NN O O
date NN O O
have NN O O
not NN O O
been NN O O
comparable NN O O
across NN O O
corpora NN O O
or NN O O
lexicons NN O O
. NN O O

Observing NN O O
that NN O O
the NN O O
quality NN O O
of NN O O
the NN O O
lexicon NN O O
greatly NN O O
impacts NN O O
the NN O O
accuracy NN O I-Metric
that NN O O
can NN O O
be NN O O
achieved NN O O
by NN O O
the NN O O
algorithms NN O I-Generic
, NN O O
we NN O O
present NN O O
a NN O O
method NN O O
of NN O O
HMM NN O I-Method
training NN O I-Method
that NN O O
improves NN O O
accuracy NN O O
when NN O O
training NN O O
of NN O O
lexical NN O I-OtherScientificTerm
probabilities NN O I-OtherScientificTerm
is NN O O
unstable NN O O
. NN O O

Finally NN O O
, NN O O
we NN O O
show NN O O
how NN O O
this NN O O
new NN O O
tagger NN O I-Generic
achieves NN O O
state-of-the-art NN O O
results NN O O
in NN O O
a NN O O
supervised NN O I-Task
, NN O I-Task
non-training NN O I-Task
intensive NN O I-Task
framework NN O I-Task
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
a NN O O
family NN O O
of NN O O
non-uniform NN O I-Method
sampling NN O I-Method
strategies NN O I-Method
to NN O O
provably NN O O
speed NN O O
up NN O O
a NN O O
class NN O O
of NN O O
stochastic NN O I-Method
optimization NN O I-Method
algorithms NN O I-Method
with NN O O
linear NN O I-OtherScientificTerm
convergence NN O I-OtherScientificTerm
including NN O O
Stochastic NN O I-Method
Variance NN O I-Method
Reduced NN O I-Method
Gradient NN O I-Method
-LRB- NN O I-Method
SVRG NN O I-Method
-RRB- NN O I-Method
and NN O O
Stochastic NN O I-Method
Dual NN O I-Method
Coordinate NN O I-Method
Ascent NN O I-Method
-LRB- NN O I-Method
SDCA NN O I-Method
-RRB- NN O I-Method
. NN O O

For NN O O
a NN O O
large NN O O
family NN O O
of NN O O
penalized NN O I-Task
empirical NN O I-Task
risk NN O I-Task
minimization NN O I-Task
problems NN O I-Task
, NN O O
our NN O O
methods NN O I-Generic
exploit NN O O
data NN O I-OtherScientificTerm
dependent NN O I-OtherScientificTerm
local NN O I-OtherScientificTerm
smoothness NN O I-OtherScientificTerm
of NN O O
the NN O O
loss NN O I-OtherScientificTerm
functions NN O I-OtherScientificTerm
near NN O O
the NN O O
optimum NN O I-OtherScientificTerm
, NN O O
while NN O O
maintaining NN O O
convergence NN O I-OtherScientificTerm
guarantees NN O I-OtherScientificTerm
. NN O O

Our NN O O
bounds NN O O
are NN O O
the NN O O
first NN O O
to NN O O
quantify NN O O
the NN O O
advantage NN O O
gained NN O O
from NN O O
local NN O I-OtherScientificTerm
smoothness NN O I-OtherScientificTerm
which NN O O
are NN O O
significant NN O O
for NN O O
some NN O O
problems NN O O
significantly NN O O
better NN O O
. NN O O

Empirically NN O O
, NN O O
we NN O O
provide NN O O
thorough NN O O
numerical NN O O
results NN O O
to NN O O
back NN O O
up NN O O
our NN O O
theory NN O I-Generic
. NN O O

Additionally NN O O
we NN O O
present NN O O
algorithms NN O I-Generic
exploiting NN O O
local NN O I-OtherScientificTerm
smoothness NN O I-OtherScientificTerm
in NN O O
more NN O O
aggressive NN O O
ways NN O O
, NN O O
which NN O O
perform NN O O
even NN O O
better NN O O
in NN O O
practice NN O O
. NN O O

-DOCSTART- NN O O

Statistical NN O I-Method
language NN O I-Method
modeling NN O I-Method
remains NN O O
a NN O O
challenging NN O O
task NN O I-Generic
, NN O O
in NN O O
particular NN O O
for NN O O
morphologically NN O I-Material
rich NN O I-Material
languages NN O I-Material
. NN O O

Recently NN O O
, NN O O
new NN O O
approaches NN O I-Generic
based NN O O
on NN O O
factored NN O I-Method
language NN O I-Method
models NN O I-Method
have NN O O
been NN O O
developed NN O O
to NN O O
address NN O O
this NN O O
problem NN O O
. NN O O

These NN O O
models NN O I-Generic
provide NN O O
principled NN O O
ways NN O O
of NN O O
including NN O O
additional NN O O
conditioning NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
other NN O O
than NN O O
the NN O O
preceding NN O O
words NN O O
, NN O O
such NN O O
as NN O O
morphological NN O I-OtherScientificTerm
or NN O I-OtherScientificTerm
syntactic NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
. NN O O

However NN O O
, NN O O
the NN O O
number NN O O
of NN O O
possible NN O O
choices NN O O
for NN O O
model NN O I-OtherScientificTerm
parameters NN O I-OtherScientificTerm
creates NN O O
a NN O O
large NN O O
space NN O O
of NN O O
models NN O O
that NN O O
can NN O O
not NN O O
be NN O O
searched NN O O
exhaustively NN O O
. NN O O

This NN O O
paper NN O O
presents NN O O
an NN O O
entirely NN O I-Method
data-driven NN O I-Method
model NN O I-Method
selection NN O I-Method
procedure NN O I-Method
based NN O O
on NN O O
genetic NN O I-Method
search NN O I-Method
, NN O O
which NN O O
is NN O O
shown NN O O
to NN O O
outperform NN O O
both NN O O
knowledge-based NN O I-Method
and NN O I-Method
random NN O I-Method
selection NN O I-Method
procedures NN O I-Method
on NN O O
two NN O O
different NN O O
language NN O I-Task
modeling NN O I-Task
tasks NN O I-Task
-LRB- NN O O
Arabic NN O I-Material
and NN O O
Turkish NN O I-Material
-RRB- NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
address NN O O
appropriate NN O O
user NN O I-Method
modeling NN O I-Method
in NN O O
order NN O O
to NN O O
generate NN O O
cooperative NN O I-OtherScientificTerm
responses NN O I-OtherScientificTerm
to NN O O
each NN O O
user NN O O
in NN O O
spoken NN O I-Task
dialogue NN O I-Task
systems NN O I-Task
. NN O O

Unlike NN O O
previous NN O O
studies NN O I-Generic
that NN O O
focus NN O O
on NN O O
user NN O O
's NN O O
knowledge NN O O
or NN O O
typical NN O O
kinds NN O O
of NN O O
users NN O O
, NN O O
the NN O O
user NN O I-Method
model NN O I-Method
we NN O O
propose NN O O
is NN O O
more NN O O
comprehensive NN O O
. NN O O

Specifically NN O O
, NN O O
we NN O O
set NN O O
up NN O O
three NN O O
dimensions NN O O
of NN O O
user NN O I-Method
models NN O I-Method
: NN O O
skill NN O O
level NN O O
to NN O O
the NN O O
system NN O O
, NN O O
knowledge NN O O
level NN O O
on NN O O
the NN O O
target NN O O
domain NN O O
and NN O O
the NN O O
degree NN O O
of NN O O
hastiness NN O O
. NN O O

Moreover NN O O
, NN O O
the NN O O
models NN O I-Generic
are NN O O
automatically NN O O
derived NN O O
by NN O O
decision NN O I-Method
tree NN O I-Method
learning NN O I-Method
using NN O O
real NN O I-Material
dialogue NN O I-Material
data NN O I-Material
collected NN O O
by NN O O
the NN O O
system NN O I-Generic
. NN O O

We NN O O
obtained NN O O
reasonable NN O O
classification NN O I-Metric
accuracy NN O I-Metric
for NN O O
all NN O O
dimensions NN O O
. NN O O

Dialogue NN O I-Method
strategies NN O I-Method
based NN O O
on NN O O
the NN O O
user NN O I-Method
modeling NN O I-Method
are NN O O
implemented NN O O
in NN O O
Kyoto NN O I-Method
city NN O I-Method
bus NN O I-Method
information NN O I-Method
system NN O I-Method
that NN O O
has NN O O
been NN O O
developed NN O O
at NN O O
our NN O O
laboratory NN O O
. NN O O

Experimental NN O O
evaluation NN O O
shows NN O O
that NN O O
the NN O O
cooperative NN O I-OtherScientificTerm
responses NN O I-OtherScientificTerm
adaptive NN O O
to NN O O
individual NN O O
users NN O O
serve NN O O
as NN O O
good NN O O
guidance NN O O
for NN O O
novice NN O O
users NN O O
without NN O O
increasing NN O O
the NN O O
dialogue NN O I-OtherScientificTerm
duration NN O I-OtherScientificTerm
for NN O O
skilled NN O O
users NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
a NN O O
novel NN O O
method NN O I-Generic
of NN O O
building NN O I-Task
polarity-tagged NN O I-Task
corpus NN O I-Task
from NN O O
HTML NN O I-Material
documents NN O I-Material
. NN O O

The NN O O
characteristics NN O O
of NN O O
this NN O O
method NN O I-Generic
is NN O O
that NN O O
it NN O I-Generic
is NN O O
fully NN O O
automatic NN O O
and NN O O
can NN O O
be NN O O
applied NN O O
to NN O O
arbitrary NN O O
HTML NN O I-Material
documents NN O I-Material
. NN O O

The NN O O
idea NN O O
behind NN O O
our NN O O
method NN O I-Generic
is NN O O
to NN O O
utilize NN O O
certain NN O O
layout NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
and NN O O
linguistic NN O I-OtherScientificTerm
pattern NN O I-OtherScientificTerm
. NN O O

By NN O O
using NN O O
them NN O I-Generic
, NN O O
we NN O O
can NN O O
automatically NN O O
extract NN O O
such NN O O
sentences NN O O
that NN O O
express NN O O
opinion NN O O
. NN O O

In NN O O
our NN O O
experiment NN O O
, NN O O
the NN O O
method NN O I-Generic
could NN O O
construct NN O O
a NN O O
corpus NN O O
consisting NN O O
of NN O O
126,610 NN O O
sentences NN O O
. NN O O

-DOCSTART- NN O O

Previous NN O O
work NN O O
has NN O O
used NN O O
monolingual NN O I-Material
parallel NN O I-Material
corpora NN O I-Material
to NN O O
extract NN O O
and NN O O
generate NN O O
paraphrases NN O I-OtherScientificTerm
. NN O O

We NN O O
show NN O O
that NN O O
this NN O O
task NN O I-Generic
can NN O O
be NN O O
done NN O O
using NN O O
bilingual NN O I-Material
parallel NN O I-Material
corpora NN O I-Material
, NN O O
a NN O O
much NN O O
more NN O O
commonly NN O O
available NN O O
resource NN O O
. NN O O

Using NN O O
alignment NN O I-Method
techniques NN O I-Method
from NN O O
phrase-based NN O I-Task
statistical NN O I-Task
machine NN O I-Task
translation NN O I-Task
, NN O O
we NN O O
show NN O O
how NN O O
paraphrases NN O I-OtherScientificTerm
in NN O O
one NN O O
language NN O O
can NN O O
be NN O O
identified NN O O
using NN O O
a NN O O
phrase NN O O
in NN O O
another NN O O
language NN O O
as NN O O
a NN O O
pivot NN O O
. NN O O

We NN O O
define NN O O
a NN O O
paraphrase NN O I-OtherScientificTerm
probability NN O I-OtherScientificTerm
that NN O O
allows NN O O
paraphrases NN O I-OtherScientificTerm
extracted NN O O
from NN O O
a NN O O
bilingual NN O I-Material
parallel NN O I-Material
corpus NN O I-Material
to NN O O
be NN O O
ranked NN O O
using NN O O
translation NN O I-OtherScientificTerm
probabilities NN O I-OtherScientificTerm
, NN O O
and NN O O
show NN O O
how NN O O
it NN O I-Generic
can NN O O
be NN O O
refined NN O O
to NN O O
take NN O O
contextual NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
into NN O O
account NN O O
. NN O O

We NN O O
evaluate NN O O
our NN O O
paraphrase NN O I-Method
extraction NN O I-Method
and NN O I-Method
ranking NN O I-Method
methods NN O I-Method
using NN O O
a NN O O
set NN O O
of NN O O
manual NN O I-Material
word NN O I-Material
alignments NN O I-Material
, NN O O
and NN O O
contrast NN O O
the NN O O
quality NN O I-Metric
with NN O O
paraphrases NN O I-OtherScientificTerm
extracted NN O O
from NN O O
automatic NN O I-OtherScientificTerm
alignments NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
an NN O O
automatic NN O O
, NN O O
essentially NN O O
domain-independent NN O I-Method
means NN O I-Method
of NN O I-Method
evaluating NN O I-Method
Spoken NN O I-Method
Language NN O I-Method
Systems NN O I-Method
-LRB- NN O I-Method
SLS NN O I-Method
-RRB- NN O I-Method
which NN O O
combines NN O O
software NN O I-Generic
we NN O O
have NN O O
developed NN O O
for NN O O
that NN O O
purpose NN O O
-LRB- NN O O
the NN O O
'' NN O O
Comparator NN O I-Method
'' NN O O
-RRB- NN O O
and NN O O
a NN O O
set NN O O
of NN O O
specifications NN O I-Generic
for NN O O
answer NN O I-OtherScientificTerm
expressions NN O I-OtherScientificTerm
-LRB- NN O O
the NN O O
'' NN O O
Common NN O I-Method
Answer NN O I-Method
Specification NN O I-Method
'' NN O O
, NN O O
or NN O O
CAS NN O I-Method
-RRB- NN O O
. NN O O

The NN O O
Comparator NN O I-Method
checks NN O O
whether NN O O
the NN O O
answer NN O O
provided NN O O
by NN O O
a NN O O
SLS NN O I-Method
accords NN O O
with NN O O
a NN O O
canonical NN O O
answer NN O O
, NN O O
returning NN O O
either NN O O
true NN O O
or NN O O
false NN O O
. NN O O

The NN O O
Common NN O I-Method
Answer NN O I-Method
Specification NN O I-Method
determines NN O O
the NN O O
syntax NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
answer NN O I-OtherScientificTerm
expressions NN O I-OtherScientificTerm
, NN O O
the NN O O
minimal NN O O
content NN O O
that NN O O
must NN O O
be NN O O
included NN O O
in NN O O
them NN O O
, NN O O
the NN O O
data NN O O
to NN O O
be NN O O
included NN O O
in NN O O
and NN O O
excluded NN O O
from NN O O
test NN O O
corpora NN O O
, NN O O
and NN O O
the NN O O
procedures NN O O
used NN O O
by NN O O
the NN O O
Comparator NN O I-Method
. NN O O

Though NN O O
some NN O O
details NN O O
of NN O O
the NN O O
CAS NN O I-OtherScientificTerm
are NN O O
particular NN O O
to NN O O
individual NN O O
domains NN O O
, NN O O
the NN O O
Comparator NN O I-Method
software NN O I-Method
is NN O O
domain-independent NN O O
, NN O O
as NN O O
is NN O O
the NN O O
CAS NN O I-Method
approach NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
an NN O O
unsupervised NN O I-Method
learning NN O I-Method
method NN O I-Method
for NN O O
associative NN O I-OtherScientificTerm
relationships NN O I-OtherScientificTerm
between NN O I-OtherScientificTerm
verb NN O I-OtherScientificTerm
phrases NN O I-OtherScientificTerm
, NN O O
which NN O O
is NN O O
important NN O O
in NN O O
developing NN O O
reliable NN O O
Q&A NN O I-Task
systems NN O I-Task
. NN O O

Consider NN O O
the NN O O
situation NN O O
that NN O O
a NN O O
user NN O O
gives NN O O
a NN O O
query NN O O
`` NN O O
How NN O O
much NN O O
petrol NN O O
was NN O O
imported NN O O
to NN O O
Japan NN O O
from NN O O
Saudi NN O O
Arabia NN O O
? NN O O
'' NN O O

to NN O O
a NN O O
Q&A NN O I-Task
system NN O I-Task
, NN O O
but NN O O
the NN O O
text NN O O
given NN O O
to NN O O
the NN O O
system NN O O
includes NN O O
only NN O O
the NN O O
description NN O O
`` NN O O
X NN O O
tonnes NN O O
of NN O O
petrol NN O O
was NN O O
conveyed NN O O
to NN O O
Japan NN O O
from NN O O
Saudi NN O O
Arabia NN O O
'' NN O O
. NN O O

We NN O O
think NN O O
that NN O O
the NN O O
description NN O O
is NN O O
a NN O O
good NN O O
clue NN O O
to NN O O
find NN O O
the NN O O
answer NN O O
for NN O O
our NN O O
query NN O O
, NN O O
`` NN O O
X NN O O
tonnes NN O O
'' NN O O
. NN O O

But NN O O
there NN O O
is NN O O
no NN O O
large-scale NN O I-Material
database NN O I-Material
that NN O O
provides NN O O
the NN O O
associative NN O I-OtherScientificTerm
relationship NN O I-OtherScientificTerm
between NN O O
`` NN O O
imported NN O O
'' NN O O
and NN O O
`` NN O O
conveyed NN O O
'' NN O O
. NN O O

Our NN O O
aim NN O O
is NN O O
to NN O O
develop NN O O
an NN O O
unsupervised NN O I-Method
learning NN O I-Method
method NN O I-Method
that NN O O
can NN O O
obtain NN O O
such NN O O
an NN O O
associative NN O I-OtherScientificTerm
relationship NN O I-OtherScientificTerm
, NN O O
which NN O O
we NN O O
call NN O O
scenario NN O I-OtherScientificTerm
consistency NN O I-OtherScientificTerm
. NN O O

The NN O O
method NN O I-Generic
we NN O O
are NN O O
currently NN O O
working NN O O
on NN O O
uses NN O O
an NN O O
expectation-maximization NN O I-Method
-LRB- NN O I-Method
EM NN O I-Method
-RRB- NN O I-Method
based NN O I-Method
word-clustering NN O I-Method
algorithm NN O I-Method
, NN O O
and NN O O
we NN O O
have NN O O
evaluated NN O O
the NN O O
effectiveness NN O O
of NN O O
this NN O O
method NN O I-Generic
using NN O O
Japanese NN O I-Material
verb NN O I-Material
phrases NN O I-Material
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
the NN O O
use NN O O
of NN O O
text NN O I-Material
data NN O I-Material
scraped NN O O
from NN O O
the NN O O
web NN O I-Material
to NN O O
augment NN O O
language NN O I-Method
models NN O I-Method
for NN O O
Automatic NN O I-Task
Speech NN O I-Task
Recognition NN O I-Task
and NN O O
Keyword NN O I-Task
Search NN O I-Task
for NN O O
Low NN O I-Material
Resource NN O I-Material
Languages NN O I-Material
. NN O O

We NN O O
scrape NN O O
text NN O O
from NN O O
multiple NN O O
genres NN O I-Generic
including NN O O
blogs NN O I-Material
, NN O O
online NN O I-Material
news NN O I-Material
, NN O O
translated NN O I-Material
TED NN O I-Material
talks NN O I-Material
, NN O O
and NN O O
subtitles NN O I-Material
. NN O O

Using NN O O
linearly NN O I-Method
interpolated NN O I-Method
language NN O I-Method
models NN O I-Method
, NN O O
we NN O O
find NN O O
that NN O O
blogs NN O I-Material
and NN O O
movie NN O I-Material
subtitles NN O I-Material
are NN O O
more NN O O
relevant NN O O
for NN O O
language NN O I-Method
modeling NN O I-Method
of NN O I-Method
conversational NN O I-Method
telephone NN O I-Method
speech NN O I-Method
and NN O O
obtain NN O O
large NN O O
reductions NN O O
in NN O O
out-of-vocabulary NN O I-OtherScientificTerm
keywords NN O I-OtherScientificTerm
. NN O O

Furthermore NN O O
, NN O O
we NN O O
show NN O O
that NN O O
the NN O O
web NN O I-Material
data NN O I-Material
can NN O O
improve NN O O
Term NN O I-Metric
Error NN O I-Metric
Rate NN O I-Metric
Performance NN O I-Metric
by NN O O
3.8 NN O O
% NN O O
absolute NN O O
and NN O O
Maximum NN O I-Metric
Term-Weighted NN O I-Metric
Value NN O I-Metric
in NN O O
Keyword NN O I-Task
Search NN O I-Task
by NN O O
0.0076-0 NN O O
.1059 NN O O
absolute NN O O
points NN O O
. NN O O

Much NN O O
of NN O O
the NN O O
gain NN O O
comes NN O O
from NN O O
the NN O O
reduction NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
out-of-vocabulary NN O I-OtherScientificTerm
items NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Pipelined NN O I-Method
Natural NN O I-Method
Language NN O I-Method
Generation NN O I-Method
-LRB- NN O I-Method
NLG NN O I-Method
-RRB- NN O I-Method
systems NN O I-Method
have NN O O
grown NN O O
increasingly NN O O
complex NN O O
as NN O O
architectural NN O I-Method
modules NN O I-Method
were NN O O
added NN O O
to NN O O
support NN O O
language NN O I-OtherScientificTerm
functionalities NN O I-OtherScientificTerm
such NN O O
as NN O O
referring NN O I-OtherScientificTerm
expressions NN O I-OtherScientificTerm
, NN O O
lexical NN O I-OtherScientificTerm
choice NN O I-OtherScientificTerm
, NN O O
and NN O O
revision NN O I-OtherScientificTerm
. NN O O

This NN O O
has NN O O
given NN O O
rise NN O O
to NN O O
discussions NN O O
about NN O O
the NN O O
relative NN O O
placement NN O O
of NN O O
these NN O O
new NN O O
modules NN O I-Generic
in NN O O
the NN O O
overall NN O I-Generic
architecture NN O I-Generic
. NN O O

Recent NN O O
work NN O O
on NN O O
another NN O O
aspect NN O O
of NN O O
multi-paragraph NN O I-Material
text NN O I-Material
, NN O O
discourse NN O I-OtherScientificTerm
markers NN O I-OtherScientificTerm
, NN O O
indicates NN O O
it NN O O
is NN O O
time NN O O
to NN O O
consider NN O O
where NN O O
a NN O O
discourse NN O I-Method
marker NN O I-Method
insertion NN O I-Method
algorithm NN O I-Method
fits NN O O
in NN O O
. NN O O

We NN O O
present NN O O
examples NN O O
which NN O O
suggest NN O O
that NN O O
in NN O O
a NN O O
pipelined NN O I-Method
NLG NN O I-Method
architecture NN O I-Method
, NN O O
the NN O O
best NN O O
approach NN O I-Generic
is NN O O
to NN O O
strongly NN O O
tie NN O O
it NN O I-Generic
to NN O O
a NN O O
revision NN O I-Method
component NN O I-Method
. NN O O

Finally NN O O
, NN O O
we NN O O
evaluate NN O O
the NN O O
approach NN O I-Generic
in NN O O
a NN O O
working NN O O
multi-page NN O I-Task
system NN O I-Task
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
a NN O O
system NN O I-Generic
which NN O O
understands NN O O
and NN O O
conceptualizes NN O O
scenes NN O I-Task
descriptions NN O I-Task
in NN O I-Task
natural NN O I-Task
language NN O I-Task
is NN O O
presented NN O O
. NN O O

Specifically NN O O
, NN O O
the NN O O
following NN O O
components NN O I-Generic
of NN O O
the NN O O
system NN O I-Generic
are NN O O
described NN O O
: NN O O
the NN O O
syntactic NN O I-Method
analyzer NN O I-Method
, NN O O
based NN O O
on NN O O
a NN O O
Procedural NN O I-Method
Systemic NN O I-Method
Grammar NN O I-Method
, NN O O
the NN O O
semantic NN O I-Method
analyzer NN O I-Method
relying NN O O
on NN O O
the NN O O
Conceptual NN O I-Method
Dependency NN O I-Method
Theory NN O I-Method
, NN O O
and NN O O
the NN O O
dictionary NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
article NN O O
considers NN O O
approaches NN O I-Generic
which NN O O
rerank NN O O
the NN O O
output NN O O
of NN O O
an NN O O
existing NN O O
probabilistic NN O I-Method
parser NN O I-Method
. NN O O

The NN O O
base NN O O
parser NN O I-Method
produces NN O O
a NN O O
set NN O O
of NN O O
candidate NN O I-OtherScientificTerm
parses NN O I-OtherScientificTerm
for NN O O
each NN O O
input NN O O
sentence NN O O
, NN O O
with NN O O
associated NN O O
probabilities NN O O
that NN O O
define NN O O
an NN O O
initial NN O O
ranking NN O I-OtherScientificTerm
of NN O O
these NN O O
parses NN O I-Generic
. NN O O

A NN O O
second NN O O
model NN O I-Generic
then NN O O
attempts NN O O
to NN O O
improve NN O O
upon NN O O
this NN O O
initial NN O O
ranking NN O I-OtherScientificTerm
, NN O O
using NN O O
additional NN O O
features NN O I-OtherScientificTerm
of NN O O
the NN O O
tree NN O I-OtherScientificTerm
as NN O O
evidence NN O O
. NN O O

The NN O O
strength NN O O
of NN O O
our NN O O
approach NN O I-Generic
is NN O O
that NN O O
it NN O I-Generic
allows NN O O
a NN O O
tree NN O I-OtherScientificTerm
to NN O O
be NN O O
represented NN O O
as NN O O
an NN O O
arbitrary NN O O
set NN O O
of NN O O
features NN O I-OtherScientificTerm
, NN O O
without NN O O
concerns NN O O
about NN O O
how NN O O
these NN O O
features NN O I-OtherScientificTerm
interact NN O O
or NN O O
overlap NN O O
and NN O O
without NN O O
the NN O O
need NN O O
to NN O O
define NN O O
a NN O O
derivation NN O I-OtherScientificTerm
or NN O O
a NN O O
generative NN O I-Method
model NN O I-Method
which NN O O
takes NN O O
these NN O O
features NN O I-OtherScientificTerm
into NN O O
account NN O O
. NN O O

We NN O O
introduce NN O O
a NN O O
new NN O O
method NN O I-Generic
for NN O O
the NN O O
reranking NN O I-Task
task NN O I-Task
, NN O O
based NN O O
on NN O O
the NN O O
boosting NN O I-Method
approach NN O I-Method
to NN O O
ranking NN O I-Task
problems NN O I-Task
described NN O O
in NN O O
Freund NN O O
et NN O O
al. NN O O
-LRB- NN O O
1998 NN O O
-RRB- NN O O
. NN O O

We NN O O
apply NN O O
the NN O O
boosting NN O I-Method
method NN O I-Method
to NN O O
parsing NN O I-Task
the NN O O
Wall NN O I-Material
Street NN O I-Material
Journal NN O I-Material
treebank NN O I-Material
. NN O O

The NN O O
method NN O I-Generic
combined NN O O
the NN O O
log-likelihood NN O I-OtherScientificTerm
under NN O O
a NN O O
baseline NN O I-Method
model NN O I-Method
-LRB- NN O O
that NN O O
of NN O O
Collins NN O O
-LSB- NN O O
1999 NN O O
-RSB- NN O O
-RRB- NN O O
with NN O O
evidence NN O O
from NN O O
an NN O O
additional NN O O
500,000 NN O O
features NN O I-OtherScientificTerm
over NN O O
parse NN O I-OtherScientificTerm
trees NN O I-OtherScientificTerm
that NN O O
were NN O O
not NN O O
included NN O O
in NN O O
the NN O O
original NN O O
model NN O I-Generic
. NN O O

The NN O O
new NN O O
model NN O I-Generic
achieved NN O O
89.75 NN O O
% NN O O
F-measure NN O I-Metric
, NN O O
a NN O O
13 NN O O
% NN O O
relative NN O O
decrease NN O O
in NN O O
F-measure NN O I-Metric
error NN O O
over NN O O
the NN O O
baseline NN O I-Generic
model NN O I-Generic
's NN O O
score NN O O
of NN O O
88.2 NN O O
% NN O O
. NN O O

The NN O O
article NN O O
also NN O O
introduces NN O O
a NN O O
new NN O O
algorithm NN O I-Generic
for NN O O
the NN O O
boosting NN O I-Method
approach NN O I-Method
which NN O O
takes NN O O
advantage NN O O
of NN O O
the NN O O
sparsity NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
in NN O O
the NN O O
parsing NN O I-Material
data NN O I-Material
. NN O O

Experiments NN O O
show NN O O
significant NN O O
efficiency NN O O
gains NN O O
for NN O O
the NN O O
new NN O O
algorithm NN O I-Generic
over NN O O
the NN O O
obvious NN O O
implementation NN O O
of NN O O
the NN O O
boosting NN O I-Method
approach NN O I-Method
. NN O O

We NN O O
argue NN O O
that NN O O
the NN O O
method NN O I-Generic
is NN O O
an NN O O
appealing NN O O
alternative NN O O
- NN O O
in NN O O
terms NN O O
of NN O O
both NN O O
simplicity NN O O
and NN O O
efficiency NN O O
- NN O O
to NN O O
work NN O O
on NN O O
feature NN O I-Method
selection NN O I-Method
methods NN O I-Method
within NN O O
log-linear NN O I-Method
-LRB- NN O I-Method
maximum-entropy NN O I-Method
-RRB- NN O I-Method
models NN O I-Method
. NN O O

Although NN O O
the NN O O
experiments NN O O
in NN O O
this NN O O
article NN O O
are NN O O
on NN O O
natural NN O I-Task
language NN O I-Task
parsing NN O I-Task
-LRB- NN O I-Task
NLP NN O I-Task
-RRB- NN O I-Task
, NN O O
the NN O O
approach NN O O
should NN O O
be NN O O
applicable NN O O
to NN O O
many NN O O
other NN O O
NLP NN O I-Task
problems NN O I-Task
which NN O O
are NN O O
naturally NN O O
framed NN O O
as NN O O
ranking NN O I-Task
tasks NN O I-Task
, NN O O
for NN O O
example NN O O
, NN O O
speech NN O I-Task
recognition NN O I-Task
, NN O O
machine NN O I-Task
translation NN O I-Task
, NN O O
or NN O O
natural NN O I-Task
language NN O I-Task
generation NN O I-Task
. NN O O

-DOCSTART- NN O O

A NN O O
model NN O I-Generic
is NN O O
presented NN O O
to NN O O
characterize NN O O
the NN O O
class NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
languages NN O I-OtherScientificTerm
obtained NN O O
by NN O O
adding NN O O
reduplication NN O I-OtherScientificTerm
to NN O O
context-free NN O I-Material
languages NN O I-Material
. NN O O

The NN O O
model NN O I-Method
is NN O O
a NN O O
pushdown NN O I-Method
automaton NN O I-Method
augmented NN O O
with NN O O
the NN O O
ability NN O O
to NN O O
check NN O O
reduplication NN O I-OtherScientificTerm
by NN O O
using NN O O
the NN O O
stack NN O I-OtherScientificTerm
in NN O O
a NN O O
new NN O O
way NN O O
. NN O O

The NN O O
class NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
languages NN O I-OtherScientificTerm
generated NN O O
is NN O O
shown NN O O
to NN O O
lie NN O O
strictly NN O O
between NN O O
the NN O O
context-free NN O I-Material
languages NN O I-Material
and NN O O
the NN O O
indexed NN O I-Material
languages NN O I-Material
. NN O O

The NN O O
model NN O I-Generic
appears NN O O
capable NN O O
of NN O O
accommodating NN O O
the NN O O
sort NN O O
of NN O O
reduplications NN O I-OtherScientificTerm
that NN O O
have NN O O
been NN O O
observed NN O O
to NN O O
occur NN O O
in NN O O
natural NN O I-Material
languages NN O I-Material
, NN O O
but NN O O
it NN O I-Generic
excludes NN O O
many NN O O
of NN O O
the NN O O
unnatural NN O O
constructions NN O O
that NN O O
other NN O O
formal NN O I-Method
models NN O I-Method
have NN O O
permitted NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
an NN O O
image NN O I-Method
set NN O I-Method
classification NN O I-Method
algorithm NN O I-Method
based NN O O
on NN O O
unsupervised NN O I-Method
clustering NN O I-Method
of NN O O
labeled NN O I-Material
training NN O I-Material
and NN O I-Material
unla-beled NN O I-Material
test NN O I-Material
data NN O I-Material
where NN O O
labels NN O O
are NN O O
only NN O O
used NN O O
in NN O O
the NN O O
stopping NN O I-OtherScientificTerm
criterion NN O I-OtherScientificTerm
. NN O O

The NN O O
probability NN O I-OtherScientificTerm
distribution NN O I-OtherScientificTerm
of NN O O
each NN O O
class NN O O
over NN O O
the NN O O
set NN O O
of NN O O
clusters NN O I-OtherScientificTerm
is NN O O
used NN O O
to NN O O
define NN O O
a NN O O
true NN O O
set NN O I-Metric
based NN O I-Metric
similarity NN O I-Metric
measure NN O I-Metric
. NN O O

To NN O O
this NN O O
end NN O O
, NN O O
we NN O O
propose NN O O
an NN O O
iterative NN O I-Method
sparse NN O I-Method
spectral NN O I-Method
clustering NN O I-Method
algorithm NN O I-Method
. NN O O

In NN O O
each NN O O
iteration NN O O
, NN O O
a NN O O
proximity NN O I-OtherScientificTerm
matrix NN O I-OtherScientificTerm
is NN O O
efficiently NN O O
recomputed NN O O
to NN O O
better NN O O
represent NN O O
the NN O O
local NN O I-OtherScientificTerm
subspace NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
. NN O O

Initial NN O I-OtherScientificTerm
clusters NN O I-OtherScientificTerm
capture NN O O
the NN O O
global NN O I-OtherScientificTerm
data NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
and NN O O
finer NN O I-OtherScientificTerm
clusters NN O I-OtherScientificTerm
at NN O O
the NN O O
later NN O O
stages NN O O
capture NN O O
the NN O O
subtle NN O I-OtherScientificTerm
class NN O I-OtherScientificTerm
differences NN O I-OtherScientificTerm
not NN O O
visible NN O O
at NN O O
the NN O O
global NN O I-OtherScientificTerm
scale NN O I-OtherScientificTerm
. NN O O

Image NN O I-Material
sets NN O I-Material
are NN O O
compactly NN O O
represented NN O O
with NN O O
multiple NN O O
Grass-mannian NN O I-Method
manifolds NN O I-Method
which NN O O
are NN O O
subsequently NN O O
embedded NN O O
in NN O O
Euclidean NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
with NN O O
the NN O O
proposed NN O O
spectral NN O I-Method
clustering NN O I-Method
algorithm NN O I-Method
. NN O O

We NN O O
also NN O O
propose NN O O
an NN O O
efficient NN O O
eigenvector NN O I-Method
solver NN O I-Method
which NN O O
not NN O O
only NN O O
reduces NN O O
the NN O O
computational NN O I-Metric
cost NN O I-Metric
of NN O O
spectral NN O I-Method
clustering NN O I-Method
by NN O O
many NN O O
folds NN O O
but NN O O
also NN O O
improves NN O O
the NN O O
clustering NN O I-Metric
quality NN O I-Metric
and NN O O
final NN O O
classification NN O I-Metric
results NN O I-Metric
. NN O O

Experiments NN O O
on NN O O
five NN O O
standard NN O O
datasets NN O I-Material
and NN O O
comparison NN O O
with NN O O
seven NN O O
existing NN O O
techniques NN O O
show NN O O
the NN O O
efficacy NN O O
of NN O O
our NN O O
algorithm NN O I-Generic
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
investigates NN O O
some NN O O
computational NN O I-OtherScientificTerm
problems NN O I-OtherScientificTerm
associated NN O O
with NN O O
probabilistic NN O I-Method
translation NN O I-Method
models NN O I-Method
that NN O O
have NN O O
recently NN O O
been NN O O
adopted NN O O
in NN O O
the NN O O
literature NN O O
on NN O O
machine NN O I-Task
translation NN O I-Task
. NN O O

These NN O O
models NN O I-Generic
can NN O O
be NN O O
viewed NN O O
as NN O O
pairs NN O O
of NN O O
probabilistic NN O I-Method
context-free NN O I-Method
grammars NN O I-Method
working NN O O
in NN O O
a NN O O
` NN O O
synchronous NN O O
' NN O O
way NN O O
. NN O O

Two NN O O
hardness NN O O
results NN O O
for NN O O
the NN O O
class NN O O
NP NN O O
are NN O O
reported NN O O
, NN O O
along NN O O
with NN O O
an NN O O
exponential NN O I-OtherScientificTerm
time NN O I-OtherScientificTerm
lower-bound NN O I-OtherScientificTerm
for NN O O
certain NN O O
classes NN O O
of NN O O
algorithms NN O O
that NN O O
are NN O O
currently NN O O
used NN O O
in NN O O
the NN O O
literature NN O O
. NN O O

-DOCSTART- NN O O

Active NN O I-Method
shape NN O I-Method
models NN O I-Method
are NN O O
a NN O O
powerful NN O O
and NN O O
widely NN O O
used NN O O
tool NN O O
to NN O O
interpret NN O O
complex NN O I-Material
image NN O I-Material
data NN O I-Material
. NN O O

By NN O O
building NN O O
models NN O I-Method
of NN O I-Method
shape NN O I-Method
variation NN O I-Method
they NN O O
enable NN O O
search NN O I-Method
algorithms NN O I-Method
to NN O O
use NN O O
a NN O O
pri-ori NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
in NN O O
an NN O O
efficient NN O O
and NN O O
gainful NN O O
way NN O O
. NN O O

However NN O O
, NN O O
due NN O O
to NN O O
the NN O O
linearity NN O I-OtherScientificTerm
of NN O O
PCA NN O I-Method
, NN O O
non-linearities NN O I-OtherScientificTerm
like NN O O
rotations NN O I-OtherScientificTerm
or NN O O
independently NN O O
moving NN O O
sub-parts NN O O
in NN O O
the NN O O
data NN O O
can NN O O
deteriorate NN O O
the NN O O
resulting NN O O
model NN O O
considerably NN O O
. NN O O

Although NN O O
non-linear NN O I-Method
extensions NN O I-Method
of NN O I-Method
active NN O I-Method
shape NN O I-Method
models NN O I-Method
have NN O O
been NN O O
proposed NN O O
and NN O O
application NN O O
specific NN O O
solutions NN O O
have NN O O
been NN O O
used NN O O
, NN O O
they NN O O
still NN O O
need NN O O
a NN O O
certain NN O O
amount NN O O
of NN O O
user NN O I-OtherScientificTerm
interaction NN O I-OtherScientificTerm
during NN O O
model NN O O
building NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
the NN O O
task NN O O
of NN O O
build-ing/choosing NN O I-Task
optimal NN O I-Task
models NN O I-Task
is NN O O
tackled NN O O
in NN O O
a NN O O
more NN O O
generic NN O O
information NN O O
theoretic NN O O
fashion NN O O
. NN O O

In NN O O
particular NN O O
, NN O O
we NN O O
propose NN O O
an NN O O
algorithm NN O I-Generic
based NN O O
on NN O O
the NN O O
minimum NN O I-Method
description NN O I-Method
length NN O I-Method
principle NN O I-Method
to NN O O
find NN O O
an NN O O
optimal NN O O
subdivision NN O O
of NN O O
the NN O O
data NN O O
into NN O O
sub-parts NN O O
, NN O O
each NN O O
adequate NN O O
for NN O O
linear NN O I-Method
modeling NN O I-Method
. NN O O

This NN O O
results NN O O
in NN O O
an NN O O
overall NN O O
more NN O O
compact NN O O
model NN O O
configuration NN O O
. NN O O

Which NN O O
in NN O O
turn NN O O
leads NN O O
to NN O O
a NN O O
better NN O O
model NN O I-Generic
in NN O O
terms NN O O
of NN O O
modes NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
variations NN O I-OtherScientificTerm
. NN O O

The NN O O
proposed NN O O
method NN O I-Generic
is NN O O
evaluated NN O O
on NN O O
synthetic NN O I-Material
data NN O I-Material
, NN O O
medical NN O I-Material
images NN O I-Material
and NN O O
hand NN O I-Material
contours NN O I-Material
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
a NN O O
set NN O O
of NN O O
experiments NN O O
to NN O O
explore NN O O
statistical NN O I-Method
techniques NN O I-Method
for NN O O
ranking NN O I-Task
and NN O O
selecting NN O O
the NN O O
best NN O O
translations NN O O
in NN O O
a NN O O
graph NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
translation NN O I-OtherScientificTerm
hypotheses NN O I-OtherScientificTerm
. NN O O

In NN O O
a NN O O
previous NN O O
paper NN O O
-LRB- NN O O
Carl NN O O
, NN O O
2007 NN O O
-RRB- NN O O
we NN O O
have NN O O
described NN O O
how NN O O
the NN O O
hypotheses NN O I-OtherScientificTerm
graph NN O I-OtherScientificTerm
is NN O O
generated NN O O
through NN O O
shallow NN O I-Method
mapping NN O I-Method
and NN O O
permutation NN O I-Method
rules NN O I-Method
. NN O O

We NN O O
have NN O O
given NN O O
examples NN O O
of NN O O
its NN O O
nodes NN O I-OtherScientificTerm
consisting NN O O
of NN O O
vectors NN O I-OtherScientificTerm
representing NN O I-OtherScientificTerm
morpho-syntactic NN O I-OtherScientificTerm
properties NN O I-OtherScientificTerm
of NN O O
words NN O O
and NN O O
phrases NN O O
. NN O O

This NN O O
paper NN O O
describes NN O O
a NN O O
number NN O O
of NN O O
methods NN O I-Generic
for NN O O
elaborating NN O O
statistical NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
functions NN O I-OtherScientificTerm
from NN O O
some NN O O
of NN O O
the NN O O
vector NN O I-Method
components NN O I-Method
. NN O O

The NN O O
feature NN O I-OtherScientificTerm
functions NN O I-OtherScientificTerm
are NN O O
trained NN O O
off-line NN O O
on NN O O
different NN O O
types NN O O
of NN O O
text NN O O
and NN O O
their NN O O
log-linear NN O I-OtherScientificTerm
combination NN O I-OtherScientificTerm
is NN O O
then NN O O
used NN O O
to NN O O
retrieve NN O O
the NN O O
best NN O O
M NN O O
translation NN O I-OtherScientificTerm
paths NN O I-OtherScientificTerm
in NN O O
the NN O O
graph NN O I-OtherScientificTerm
. NN O O

We NN O O
compare NN O O
two NN O O
language NN O I-Method
modelling NN O I-Method
toolkits NN O I-Method
, NN O O
the NN O O
CMU NN O I-Method
and NN O I-Method
the NN O I-Method
SRI NN O I-Method
toolkit NN O I-Method
and NN O O
arrive NN O O
at NN O O
three NN O O
results NN O O
: NN O O
1 NN O O
-RRB- NN O O
word-lemma NN O I-Method
based NN O I-Method
feature NN O I-Method
function NN O I-Method
models NN O I-Method
produce NN O O
better NN O O
results NN O O
than NN O O
token-based NN O I-Method
models NN O I-Method
, NN O O
2 NN O O
-RRB- NN O O
adding NN O O
a NN O O
PoS-tag NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
function NN O I-OtherScientificTerm
to NN O O
the NN O O
word-lemma NN O I-Method
model NN O I-Method
improves NN O O
the NN O O
output NN O O
and NN O O
3 NN O O
-RRB- NN O O
weights NN O O
for NN O O
lexical NN O I-Task
translations NN O I-Task
are NN O O
suitable NN O O
if NN O O
the NN O O
training NN O O
material NN O O
is NN O O
similar NN O O
to NN O O
the NN O O
texts NN O O
to NN O O
be NN O O
translated NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
specialized NN O O
editor NN O I-Method
for NN O O
a NN O O
highly NN O O
structured NN O I-Material
dictionary NN O I-Material
. NN O O

The NN O O
basic NN O O
goal NN O O
in NN O O
building NN O O
that NN O O
editor NN O I-Method
was NN O O
to NN O O
provide NN O O
an NN O O
adequate NN O O
tool NN O O
to NN O O
help NN O O
lexicologists NN O O
produce NN O O
a NN O O
valid NN O O
and NN O O
coherent NN O O
dictionary NN O I-Generic
on NN O O
the NN O O
basis NN O O
of NN O O
a NN O O
linguistic NN O I-Method
theory NN O I-Method
. NN O O

If NN O O
we NN O O
want NN O O
valuable NN O O
lexicons NN O O
and NN O O
grammars NN O O
to NN O O
achieve NN O O
complex NN O O
natural NN O I-Task
language NN O I-Task
processing NN O I-Task
, NN O O
we NN O O
must NN O O
provide NN O O
very NN O O
powerful NN O O
tools NN O O
to NN O O
help NN O O
create NN O O
and NN O O
ensure NN O O
the NN O O
validity NN O O
of NN O O
such NN O O
complex NN O O
linguistic NN O I-Material
databases NN O I-Material
. NN O O

Our NN O O
most NN O O
important NN O O
task NN O O
in NN O O
building NN O O
the NN O O
editor NN O I-Method
was NN O O
to NN O O
define NN O O
a NN O O
set NN O O
of NN O O
coherence NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
that NN O O
could NN O O
be NN O O
computationally NN O O
applied NN O O
to NN O O
ensure NN O O
the NN O O
validity NN O O
of NN O O
lexical NN O O
entries NN O O
. NN O O

A NN O O
customized NN O O
interface NN O O
for NN O O
browsing NN O O
and NN O O
editing NN O O
was NN O O
also NN O O
designed NN O O
and NN O O
implemented NN O O
. NN O O

-DOCSTART- NN O O

Existing NN O O
techniques NN O O
extract NN O O
term NN O O
candidates NN O O
by NN O O
looking NN O O
for NN O O
internal NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
contextual NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
associated NN O O
with NN O O
domain NN O I-OtherScientificTerm
specific NN O I-OtherScientificTerm
terms NN O I-OtherScientificTerm
. NN O O

The NN O O
algorithms NN O O
always NN O O
face NN O O
the NN O O
dilemma NN O O
that NN O O
fewer NN O O
features NN O I-OtherScientificTerm
are NN O O
not NN O O
enough NN O O
to NN O O
distinguish NN O O
terms NN O O
from NN O O
non-terms NN O O
whereas NN O O
more NN O O
features NN O I-OtherScientificTerm
lead NN O O
to NN O O
more NN O O
conflicts NN O O
among NN O O
selected NN O O
features NN O O
. NN O O

This NN O O
paper NN O O
presents NN O O
a NN O O
novel NN O O
approach NN O I-Generic
for NN O O
term NN O I-Task
extraction NN O I-Task
based NN O O
on NN O O
delimiters NN O I-OtherScientificTerm
which NN O O
are NN O O
much NN O O
more NN O O
stable NN O O
and NN O O
domain NN O O
independent NN O O
. NN O O

The NN O O
proposed NN O O
approach NN O I-Generic
is NN O O
not NN O O
as NN O O
sensitive NN O O
to NN O O
term NN O I-OtherScientificTerm
frequency NN O I-OtherScientificTerm
as NN O O
that NN O O
of NN O O
previous NN O O
works NN O O
. NN O O

This NN O O
approach NN O I-Generic
has NN O O
no NN O O
strict NN O O
limit NN O O
or NN O O
hard NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
and NN O O
thus NN O O
they NN O O
can NN O O
deal NN O O
with NN O O
all NN O O
kinds NN O O
of NN O O
terms NN O O
. NN O O

It NN O O
also NN O O
requires NN O O
no NN O O
prior NN O O
domain NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
and NN O O
no NN O O
additional NN O O
training NN O O
to NN O O
adapt NN O O
to NN O O
new NN O O
domains NN O O
. NN O O

Consequently NN O O
, NN O O
the NN O O
proposed NN O O
approach NN O I-Generic
can NN O O
be NN O O
applied NN O O
to NN O O
different NN O O
domains NN O O
easily NN O O
and NN O O
it NN O I-Generic
is NN O O
especially NN O O
useful NN O O
for NN O O
resource-limited NN O I-Material
domains NN O I-Material
. NN O O

Evaluations NN O I-Generic
conducted NN O O
on NN O O
two NN O O
different NN O O
domains NN O O
for NN O O
Chinese NN O I-Task
term NN O I-Task
extraction NN O I-Task
show NN O O
significant NN O O
improvements NN O O
over NN O O
existing NN O O
techniques NN O O
which NN O O
verifies NN O O
its NN O O
efficiency NN O O
and NN O O
domain NN O O
independent NN O O
nature NN O O
. NN O O

Experiments NN O O
on NN O O
new NN O I-Task
term NN O I-Task
extraction NN O I-Task
indicate NN O O
that NN O O
the NN O O
proposed NN O O
approach NN O I-Generic
can NN O O
also NN O O
serve NN O O
as NN O O
an NN O O
effective NN O O
tool NN O I-Generic
for NN O O
domain NN O I-Task
lexicon NN O I-Task
expansion NN O I-Task
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
a NN O O
method NN O I-Generic
for NN O O
identifying NN O O
systematic NN O I-OtherScientificTerm
patterns NN O I-OtherScientificTerm
in NN O I-OtherScientificTerm
translation NN O I-OtherScientificTerm
data NN O I-OtherScientificTerm
using NN O O
part-of-speech NN O I-Material
tag NN O I-Material
sequences NN O I-Material
. NN O O

We NN O O
incorporate NN O O
this NN O O
analysis NN O I-Generic
into NN O O
a NN O O
diagnostic NN O I-Method
tool NN O I-Method
intended NN O O
for NN O O
developers NN O O
of NN O O
machine NN O I-Method
translation NN O I-Method
systems NN O I-Method
, NN O O
and NN O O
demonstrate NN O O
how NN O O
our NN O O
application NN O I-Generic
can NN O O
be NN O O
used NN O O
by NN O O
developers NN O O
to NN O O
explore NN O O
patterns NN O I-OtherScientificTerm
in NN O I-OtherScientificTerm
machine NN O I-OtherScientificTerm
translation NN O I-OtherScientificTerm
output NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

We NN O O
study NN O O
the NN O O
number NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
hidden NN O I-OtherScientificTerm
layers NN O I-OtherScientificTerm
required NN O O
by NN O O
a NN O O
multilayer NN O I-Method
neu-ral NN O I-Method
network NN O I-Method
with NN O O
threshold NN O I-OtherScientificTerm
units NN O I-OtherScientificTerm
to NN O O
compute NN O O
a NN O O
function NN O O
f NN O O
from NN O O
n NN O O
d NN O O
to NN O O
-LCB- NN O O
O NN O O
, NN O O
I NN O O
-RCB- NN O O
. NN O O

In NN O O
dimension NN O O
d NN O O
= NN O O
2 NN O O
, NN O O
Gibson NN O O
characterized NN O O
the NN O O
functions NN O O
computable NN O O
with NN O O
just NN O O
one NN O O
hidden NN O I-OtherScientificTerm
layer NN O I-OtherScientificTerm
, NN O O
under NN O O
the NN O O
assumption NN O O
that NN O O
there NN O O
is NN O O
no NN O O
`` NN O O
multiple NN O O
intersection NN O O
point NN O O
'' NN O O
and NN O O
that NN O O
f NN O O
is NN O O
only NN O O
defined NN O O
on NN O O
a NN O O
compact NN O O
set NN O O
. NN O O

We NN O O
consider NN O O
the NN O O
restriction NN O O
of NN O O
f NN O O
to NN O O
the NN O O
neighborhood NN O O
of NN O O
a NN O O
multiple NN O O
intersection NN O O
point NN O O
or NN O O
of NN O O
infinity NN O O
, NN O O
and NN O O
give NN O O
necessary NN O O
and NN O O
sufficient NN O O
conditions NN O O
for NN O O
it NN O O
to NN O O
be NN O O
locally NN O O
computable NN O O
with NN O O
one NN O O
hidden NN O I-OtherScientificTerm
layer NN O I-OtherScientificTerm
. NN O O

We NN O O
show NN O O
that NN O O
adding NN O O
these NN O O
conditions NN O O
to NN O O
Gib-son NN O O
's NN O O
assumptions NN O O
is NN O O
not NN O O
sufficient NN O O
to NN O O
ensure NN O O
global NN O I-OtherScientificTerm
computability NN O I-OtherScientificTerm
with NN O O
one NN O O
hidden NN O I-OtherScientificTerm
layer NN O I-OtherScientificTerm
, NN O O
by NN O O
exhibiting NN O O
a NN O O
new NN O O
non-local NN O I-OtherScientificTerm
configuration NN O I-OtherScientificTerm
, NN O O
the NN O O
`` NN O I-OtherScientificTerm
critical NN O I-OtherScientificTerm
cycle NN O I-OtherScientificTerm
'' NN O I-OtherScientificTerm
, NN O O
which NN O O
implies NN O O
that NN O O
f NN O O
is NN O O
not NN O O
computable NN O O
with NN O O
one NN O O
hidden NN O I-OtherScientificTerm
layer NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
an NN O O
approach NN O I-Generic
to NN O O
estimate NN O O
the NN O O
intrinsic NN O I-Task
texture NN O I-Task
properties NN O I-Task
-LRB- NN O I-Task
albedo NN O I-Task
, NN O I-Task
shading NN O I-Task
, NN O I-Task
normal NN O I-Task
-RRB- NN O I-Task
of NN O I-Task
scenes NN O I-Task
from NN O O
multiple NN O I-Task
view NN O I-Task
acquisition NN O I-Task
under NN O O
unknown NN O I-OtherScientificTerm
illumination NN O I-OtherScientificTerm
conditions NN O I-OtherScientificTerm
. NN O O

We NN O O
introduce NN O O
the NN O O
concept NN O O
of NN O O
intrinsic NN O I-OtherScientificTerm
textures NN O I-OtherScientificTerm
, NN O O
which NN O O
are NN O O
pixel-resolution NN O I-OtherScientificTerm
surface NN O I-OtherScientificTerm
textures NN O I-OtherScientificTerm
representing NN O O
the NN O O
intrinsic NN O I-OtherScientificTerm
appearance NN O I-OtherScientificTerm
parameters NN O I-OtherScientificTerm
of NN O O
a NN O O
scene NN O O
. NN O O

Unlike NN O O
previous NN O O
video NN O I-Method
relighting NN O I-Method
methods NN O I-Method
, NN O O
the NN O O
approach NN O I-Generic
does NN O O
not NN O O
assume NN O O
regions NN O O
of NN O O
uniform NN O I-OtherScientificTerm
albedo NN O I-OtherScientificTerm
, NN O O
which NN O O
makes NN O O
it NN O I-Generic
applicable NN O O
to NN O O
richly NN O I-Material
textured NN O I-Material
scenes NN O I-Material
. NN O O

We NN O O
show NN O O
that NN O O
intrinsic NN O I-Method
image NN O I-Method
methods NN O I-Method
can NN O O
be NN O O
used NN O O
to NN O O
refine NN O O
an NN O O
initial NN O I-Task
, NN O I-Task
low-frequency NN O I-Task
shading NN O I-Task
estimate NN O I-Task
based NN O O
on NN O O
a NN O O
global NN O I-OtherScientificTerm
lighting NN O I-OtherScientificTerm
reconstruction NN O I-OtherScientificTerm
from NN O O
an NN O O
original NN O O
texture NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
coarse NN O I-OtherScientificTerm
scene NN O I-OtherScientificTerm
geometry NN O I-OtherScientificTerm
in NN O O
order NN O O
to NN O O
resolve NN O O
the NN O O
inherent NN O I-OtherScientificTerm
global NN O I-OtherScientificTerm
ambiguity NN O I-OtherScientificTerm
in NN O I-OtherScientificTerm
shading NN O I-OtherScientificTerm
. NN O O

The NN O O
method NN O I-Generic
is NN O O
applied NN O O
to NN O O
relight-ing NN O I-Task
of NN O I-Task
free-viewpoint NN O I-Task
rendering NN O I-Task
from NN O O
multiple NN O I-OtherScientificTerm
view NN O I-OtherScientificTerm
video NN O I-OtherScientificTerm
capture NN O I-OtherScientificTerm
. NN O O

This NN O O
demonstrates NN O O
relighting NN O I-Task
with NN O O
reproduction NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
fine NN O I-OtherScientificTerm
surface NN O I-OtherScientificTerm
detail NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Following NN O O
recent NN O O
developments NN O O
in NN O O
the NN O O
automatic NN O I-Task
evaluation NN O I-Task
of NN O I-Task
machine NN O I-Task
translation NN O I-Task
and NN O I-Task
document NN O I-Task
summarization NN O I-Task
, NN O O
we NN O O
present NN O O
a NN O O
similar NN O O
approach NN O I-Generic
, NN O O
implemented NN O O
in NN O O
a NN O O
measure NN O I-Generic
called NN O O
POURPRE NN O I-Metric
, NN O O
for NN O O
automatically NN O I-Task
evaluating NN O I-Task
answers NN O I-Task
to NN O I-Task
definition NN O I-Task
questions NN O I-Task
. NN O O

Until NN O O
now NN O O
, NN O O
the NN O O
only NN O O
way NN O O
to NN O O
assess NN O O
the NN O O
correctness NN O O
of NN O O
answers NN O O
to NN O O
such NN O O
questions NN O O
involves NN O O
manual NN O O
determination NN O O
of NN O O
whether NN O O
an NN O O
information NN O O
nugget NN O O
appears NN O O
in NN O O
a NN O O
system NN O O
's NN O O
response NN O O
. NN O O

The NN O O
lack NN O O
of NN O O
automatic NN O O
methods NN O O
for NN O O
scoring NN O O
system NN O O
output NN O O
is NN O O
an NN O O
impediment NN O O
to NN O O
progress NN O O
in NN O O
the NN O O
field NN O O
, NN O O
which NN O O
we NN O O
address NN O O
with NN O O
this NN O O
work NN O O
. NN O O

Experiments NN O O
with NN O O
the NN O O
TREC NN O I-Material
2003 NN O I-Material
and NN O I-Material
TREC NN O I-Material
2004 NN O I-Material
QA NN O I-Material
tracks NN O I-Material
indicate NN O O
that NN O O
rankings NN O I-OtherScientificTerm
produced NN O O
by NN O O
our NN O O
metric NN O I-Generic
correlate NN O O
highly NN O O
with NN O O
official NN O I-OtherScientificTerm
rankings NN O I-OtherScientificTerm
, NN O O
and NN O O
that NN O O
POURPRE NN O I-Metric
outperforms NN O O
direct NN O O
application NN O O
of NN O O
existing NN O O
metrics NN O I-Generic
. NN O O

-DOCSTART- NN O O

Recent NN O O
advances NN O O
in NN O O
Automatic NN O I-Method
Speech NN O I-Method
Recognition NN O I-Method
technology NN O I-Method
have NN O O
put NN O O
the NN O O
goal NN O O
of NN O O
naturally NN O O
sounding NN O O
dialog NN O I-Task
systems NN O I-Task
within NN O O
reach NN O O
. NN O O

However NN O O
, NN O O
the NN O O
improved NN O O
speech NN O I-Task
recognition NN O I-Task
has NN O O
brought NN O O
to NN O O
light NN O O
a NN O O
new NN O O
problem NN O O
: NN O O
as NN O O
dialog NN O I-Task
systems NN O I-Task
understand NN O O
more NN O O
of NN O O
what NN O O
the NN O O
user NN O O
tells NN O O
them NN O O
, NN O O
they NN O I-Generic
need NN O O
to NN O O
be NN O O
more NN O O
sophisticated NN O O
at NN O O
responding NN O O
to NN O O
the NN O O
user NN O O
. NN O O

The NN O O
issue NN O O
of NN O O
system NN O I-OtherScientificTerm
response NN O I-OtherScientificTerm
to NN O O
users NN O O
has NN O O
been NN O O
extensively NN O O
studied NN O O
by NN O O
the NN O O
natural NN O I-Task
language NN O I-Task
generation NN O I-Task
community NN O I-Task
, NN O O
though NN O O
rarely NN O O
in NN O O
the NN O O
context NN O O
of NN O O
dialog NN O I-Task
systems NN O I-Task
. NN O O

We NN O O
show NN O O
how NN O O
research NN O O
in NN O O
generation NN O I-Task
can NN O O
be NN O O
adapted NN O O
to NN O O
dialog NN O I-Task
systems NN O I-Task
, NN O O
and NN O O
how NN O O
the NN O O
high NN O O
cost NN O O
of NN O O
hand-crafting NN O I-Method
knowledge-based NN O I-Method
generation NN O I-Method
systems NN O I-Method
can NN O O
be NN O O
overcome NN O O
by NN O O
employing NN O O
machine NN O I-Method
learning NN O I-Method
techniques NN O I-Method
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
a NN O O
tool NN O I-Generic
, NN O O
called NN O O
ILIMP NN O I-Method
, NN O O
which NN O O
takes NN O O
as NN O O
input NN O O
a NN O O
raw NN O I-Material
text NN O I-Material
in NN O I-Material
French NN O I-Material
and NN O O
produces NN O O
as NN O O
output NN O O
the NN O O
same NN O O
text NN O O
in NN O O
which NN O O
every NN O O
occurrence NN O O
of NN O O
the NN O O
pronoun NN O O
il NN O O
is NN O O
tagged NN O O
either NN O O
with NN O O
tag NN O O
-LSB- NN O O
ANA NN O O
-RSB- NN O O
for NN O O
anaphoric NN O O
or NN O O
-LSB- NN O O
IMP NN O O
-RSB- NN O O
for NN O O
impersonal NN O O
or NN O O
expletive NN O O
. NN O O

This NN O O
tool NN O I-Generic
is NN O O
therefore NN O O
designed NN O O
to NN O O
distinguish NN O O
between NN O O
the NN O O
anaphoric NN O I-Task
occurrences NN O I-Task
of NN O I-Task
il NN O I-Task
, NN O O
for NN O O
which NN O O
an NN O O
anaphora NN O I-Method
resolution NN O I-Method
system NN O I-Method
has NN O O
to NN O O
look NN O O
for NN O O
an NN O O
antecedent NN O O
, NN O O
and NN O O
the NN O O
expletive NN O O
occurrences NN O O
of NN O O
this NN O O
pronoun NN O O
, NN O O
for NN O O
which NN O O
it NN O O
does NN O O
not NN O O
make NN O O
sense NN O O
to NN O O
look NN O O
for NN O O
an NN O O
antecedent NN O O
. NN O O

The NN O O
precision NN O I-Metric
rate NN O I-Metric
for NN O O
ILIMP NN O I-Method
is NN O O
97,5 NN O O
% NN O O
. NN O O

The NN O O
few NN O O
errors NN O O
are NN O O
analyzed NN O O
in NN O O
detail NN O O
. NN O O

Other NN O O
tasks NN O I-Generic
using NN O O
the NN O O
method NN O I-Generic
developed NN O O
for NN O O
ILIMP NN O I-Method
are NN O O
described NN O O
briefly NN O O
, NN O O
as NN O O
well NN O O
as NN O O
the NN O O
use NN O O
of NN O O
ILIMP NN O I-Method
in NN O O
a NN O O
modular NN O I-Method
syntactic NN O I-Method
analysis NN O I-Method
system NN O I-Method
. NN O O

-DOCSTART- NN O O

Speech-based NN O I-Task
depression NN O I-Task
detection NN O I-Task
has NN O O
gained NN O O
importance NN O O
in NN O O
recent NN O O
years NN O O
, NN O O
but NN O O
most NN O O
research NN O O
has NN O O
used NN O O
relatively NN O O
quiet NN O O
conditions NN O O
or NN O O
examined NN O O
a NN O O
single NN O O
corpus NN O O
per NN O O
study NN O O
. NN O O

Little NN O O
is NN O O
thus NN O O
known NN O O
about NN O O
the NN O O
robustness NN O I-Metric
of NN O O
speech NN O I-OtherScientificTerm
cues NN O I-OtherScientificTerm
in NN O O
the NN O O
wild NN O O
. NN O O

This NN O O
study NN O O
compares NN O O
the NN O O
effect NN O O
of NN O O
noise NN O I-OtherScientificTerm
and NN O O
reverberation NN O I-OtherScientificTerm
on NN O O
depression NN O I-Task
prediction NN O I-Task
using NN O O
1 NN O O
-RRB- NN O O
standard NN O O
mel-frequency NN O I-Method
cepstral NN O I-Method
coefficients NN O I-Method
-LRB- NN O I-Method
MFCCs NN O I-Method
-RRB- NN O I-Method
, NN O O
and NN O O
2 NN O O
-RRB- NN O O
features NN O I-Generic
designed NN O O
for NN O O
noise NN O I-Metric
robustness NN O I-Metric
, NN O O
damped NN O I-Method
oscillator NN O I-Method
cepstral NN O I-Method
coefficients NN O I-Method
-LRB- NN O I-Method
DOCCs NN O I-Method
-RRB- NN O I-Method
. NN O O

Data NN O O
come NN O O
from NN O O
the NN O O
2014 NN O O
AudioVisual NN O I-Material
Emotion NN O I-Material
Recognition NN O I-Material
Challenge NN O I-Material
-LRB- NN O I-Material
AVEC NN O I-Material
-RRB- NN O I-Material
. NN O O

Results NN O O
using NN O O
additive NN O I-OtherScientificTerm
noise NN O I-OtherScientificTerm
and NN O O
reverberation NN O I-OtherScientificTerm
reveal NN O O
a NN O O
consistent NN O O
pattern NN O O
of NN O O
findings NN O O
for NN O O
multiple NN O O
evaluation NN O I-Metric
metrics NN O I-Metric
under NN O O
both NN O O
matched NN O O
and NN O O
mismatched NN O O
conditions NN O O
. NN O O

First NN O O
and NN O O
most NN O O
notably NN O O
: NN O O
standard NN O O
MFCC NN O I-Method
features NN O I-Method
suffer NN O O
dramatically NN O O
under NN O O
test/train NN O O
mismatch NN O O
for NN O O
both NN O O
noise NN O I-OtherScientificTerm
and NN O O
reverberation NN O I-OtherScientificTerm
; NN O O
DOCC NN O I-Method
features NN O I-Method
are NN O O
far NN O O
more NN O O
robust NN O O
. NN O O

Second NN O O
, NN O O
including NN O O
higher-order NN O I-OtherScientificTerm
cepstral NN O I-OtherScientificTerm
coefficients NN O I-OtherScientificTerm
is NN O O
generally NN O O
beneficial NN O O
. NN O O

Third NN O O
, NN O O
artificial NN O I-Method
neural NN O I-Method
networks NN O I-Method
tend NN O O
to NN O O
outperform NN O O
support NN O I-Method
vector NN O I-Method
regression NN O I-Method
. NN O O

Fourth NN O O
, NN O O
spontaneous NN O I-Material
speech NN O I-Material
appears NN O O
to NN O O
offer NN O O
better NN O O
robustness NN O O
than NN O O
read NN O I-Material
speech NN O I-Material
. NN O O

Finally NN O O
, NN O O
a NN O O
cross-corpus NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
cross-language NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
experiment NN O I-OtherScientificTerm
reveals NN O O
better NN O O
noise NN O I-Metric
and NN O I-Metric
reverberation NN O I-Metric
robustness NN O I-Metric
for NN O O
DOCCs NN O I-Method
than NN O O
for NN O O
MFCCs NN O I-Method
. NN O O

Implications NN O O
and NN O O
future NN O O
directions NN O O
for NN O O
real-world NN O I-Task
robust NN O I-Task
depression NN O I-Task
detection NN O I-Task
are NN O O
discussed NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
document NN O I-Method
oriented NN O I-Method
preference NN O I-Method
sets NN O I-Method
-LRB- NN O I-Method
DoPS NN O I-Method
-RRB- NN O I-Method
for NN O O
the NN O O
disambiguation NN O I-Task
of NN O I-Task
the NN O I-Task
dependency NN O I-Task
structure NN O I-Task
of NN O O
sentences NN O O
. NN O O

The NN O O
DoPS NN O I-Method
system NN O I-Method
extracts NN O O
preference NN O O
knowledge NN O O
from NN O O
a NN O O
target NN O O
document NN O O
or NN O O
other NN O O
documents NN O O
automatically NN O O
. NN O O

Sentence NN O I-OtherScientificTerm
ambiguities NN O I-OtherScientificTerm
can NN O O
be NN O O
resolved NN O O
by NN O O
using NN O O
domain NN O I-OtherScientificTerm
targeted NN O I-OtherScientificTerm
preference NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
without NN O O
using NN O O
complicated NN O O
large NN O O
knowledgebases NN O I-Material
. NN O O

Implementation NN O O
and NN O O
empirical NN O O
results NN O O
are NN O O
described NN O O
for NN O O
the NN O O
the NN O O
analysis NN O O
of NN O O
dependency NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
of NN O O
Japanese NN O I-Material
patent NN O I-Material
claim NN O I-Material
sentences NN O I-Material
. NN O O

-DOCSTART- NN O O

Multimodal NN O I-Task
interfaces NN O I-Task
require NN O O
effective NN O O
parsing NN O I-Method
and NN O O
understanding NN O O
of NN O O
utterances NN O O
whose NN O O
content NN O O
is NN O O
distributed NN O O
across NN O O
multiple NN O O
input NN O O
modes NN O O
. NN O O

Johnston NN O O
1998 NN O O
presents NN O O
an NN O O
approach NN O I-Generic
in NN O O
which NN O O
strategies NN O O
for NN O O
multimodal NN O I-Task
integration NN O I-Task
are NN O O
stated NN O O
declaratively NN O O
using NN O O
a NN O O
unification-based NN O I-Method
grammar NN O I-Method
that NN O O
is NN O O
used NN O O
by NN O O
a NN O O
multidimensional NN O I-Method
chart NN O I-Method
parser NN O I-Method
to NN O O
compose NN O O
inputs NN O O
. NN O O

This NN O O
approach NN O I-Generic
is NN O O
highly NN O O
expressive NN O O
and NN O O
supports NN O O
a NN O O
broad NN O O
class NN O O
of NN O O
interfaces NN O I-OtherScientificTerm
, NN O O
but NN O O
offers NN O O
only NN O O
limited NN O O
potential NN O O
for NN O O
mutual NN O O
compensation NN O O
among NN O O
the NN O O
input NN O O
modes NN O O
, NN O O
is NN O O
subject NN O O
to NN O O
significant NN O O
concerns NN O O
in NN O O
terms NN O O
of NN O O
computational NN O I-Metric
complexity NN O I-Metric
, NN O O
and NN O O
complicates NN O O
selection NN O O
among NN O O
alternative NN O O
multimodal NN O O
interpretations NN O O
of NN O O
the NN O O
input NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
an NN O O
alternative NN O O
approach NN O I-Generic
in NN O O
which NN O O
multimodal NN O I-Task
parsing NN O I-Task
and NN O I-Task
understanding NN O I-Task
are NN O O
achieved NN O O
using NN O O
a NN O O
weighted NN O I-Method
finite-state NN O I-Method
device NN O I-Method
which NN O O
takes NN O O
speech NN O I-Material
and NN O I-Material
gesture NN O I-Material
streams NN O I-Material
as NN O O
inputs NN O O
and NN O O
outputs NN O O
their NN O O
joint NN O O
interpretation NN O O
. NN O O

This NN O O
approach NN O I-Generic
is NN O O
significantly NN O O
more NN O O
efficient NN O O
, NN O O
enables NN O O
tight-coupling NN O O
of NN O O
multimodal NN O I-Task
understanding NN O I-Task
with NN O O
speech NN O I-Task
recognition NN O I-Task
, NN O O
and NN O O
provides NN O O
a NN O O
general NN O O
probabilistic NN O O
framework NN O O
for NN O O
multimodal NN O I-Task
ambiguity NN O I-Task
resolution NN O I-Task
. NN O O

-DOCSTART- NN O O

Recently NN O O
, NN O O
we NN O O
initiated NN O O
a NN O O
project NN O O
to NN O O
develop NN O O
a NN O O
phonetically-based NN O I-Method
spoken NN O I-Method
language NN O I-Method
understanding NN O I-Method
system NN O I-Method
called NN O O
SUMMIT NN O I-Method
. NN O O

In NN O O
contrast NN O O
to NN O O
many NN O O
of NN O O
the NN O O
past NN O O
efforts NN O O
that NN O O
make NN O O
use NN O O
of NN O O
heuristic NN O I-Method
rules NN O I-Method
whose NN O O
development NN O O
requires NN O O
intense NN O O
knowledge NN O I-Method
engineering NN O I-Method
, NN O O
our NN O O
approach NN O I-Generic
attempts NN O O
to NN O O
express NN O O
the NN O O
speech NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
within NN O O
a NN O O
formal NN O O
framework NN O O
using NN O O
well-defined NN O O
mathematical NN O I-Method
tools NN O I-Method
. NN O O

In NN O O
our NN O O
system NN O I-Generic
, NN O O
features NN O I-OtherScientificTerm
and NN O O
decision NN O I-Method
strategies NN O I-Method
are NN O O
discovered NN O O
and NN O O
trained NN O O
automatically NN O O
, NN O O
using NN O O
a NN O O
large NN O O
body NN O O
of NN O O
speech NN O I-Material
data NN O I-Material
. NN O O

This NN O O
paper NN O O
describes NN O O
the NN O O
system NN O I-Generic
, NN O O
and NN O O
documents NN O O
its NN O O
current NN O O
performance NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
an NN O O
implemented NN O O
program NN O I-Generic
that NN O O
takes NN O O
a NN O O
tagged NN O I-Material
text NN O I-Material
corpus NN O I-Material
and NN O O
generates NN O O
a NN O O
partial NN O O
list NN O O
of NN O O
the NN O O
subcategorization NN O I-OtherScientificTerm
frames NN O I-OtherScientificTerm
in NN O O
which NN O O
each NN O O
verb NN O O
occurs NN O O
. NN O O

The NN O O
completeness NN O O
of NN O O
the NN O O
output NN O O
list NN O O
increases NN O O
monotonically NN O O
with NN O O
the NN O O
total NN O O
occurrences NN O O
of NN O O
each NN O O
verb NN O O
in NN O O
the NN O O
training NN O O
corpus NN O O
. NN O O

False NN O I-Metric
positive NN O I-Metric
rates NN O I-Metric
are NN O O
one NN O O
to NN O O
three NN O O
percent NN O O
. NN O O

Five NN O O
subcategorization NN O I-OtherScientificTerm
frames NN O I-OtherScientificTerm
are NN O O
currently NN O O
detected NN O O
and NN O O
we NN O O
foresee NN O O
no NN O O
impediment NN O O
to NN O O
detecting NN O O
many NN O O
more NN O O
. NN O O

Ultimately NN O O
, NN O O
we NN O O
expect NN O O
to NN O O
provide NN O O
a NN O O
large NN O O
subcategorization NN O I-Material
dictionary NN O I-Material
to NN O O
the NN O O
NLP NN O O
community NN O O
and NN O O
to NN O O
train NN O O
dictionaries NN O O
for NN O O
specific NN O O
corpora NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
a NN O O
method NN O I-Generic
for NN O O
estimating NN O O
the NN O O
relative NN O I-Task
pose NN O I-Task
of NN O I-Task
two NN O I-Task
calibrated NN O I-Task
or NN O I-Task
uncalibrated NN O I-Task
non-overlapping NN O I-Task
surveillance NN O I-Task
cameras NN O I-Task
from NN O O
observing NN O O
a NN O O
moving NN O O
object NN O O
. NN O O

We NN O O
show NN O O
how NN O O
to NN O O
tackle NN O O
the NN O O
problem NN O O
of NN O O
missing NN O I-Task
point NN O I-Task
correspondences NN O I-Task
heavily NN O O
required NN O O
by NN O O
SfM NN O I-Method
pipelines NN O I-Method
and NN O O
how NN O O
to NN O O
go NN O O
beyond NN O O
this NN O O
basic NN O O
paradigm NN O I-Generic
. NN O O

We NN O O
relax NN O O
the NN O O
non-linear NN O I-OtherScientificTerm
nature NN O I-OtherScientificTerm
of NN O O
the NN O O
problem NN O I-Generic
by NN O O
accepting NN O O
two NN O O
assumptions NN O I-Generic
which NN O O
surveillance NN O I-OtherScientificTerm
scenarios NN O I-OtherScientificTerm
offer NN O O
, NN O O
ie NN O O
. NN O O

the NN O O
presence NN O O
of NN O O
a NN O O
moving NN O O
object NN O O
and NN O O
easily NN O O
estimable NN O O
gravity NN O I-OtherScientificTerm
vector NN O I-OtherScientificTerm
. NN O O

By NN O O
those NN O O
assumptions NN O I-Generic
we NN O O
cast NN O O
the NN O O
problem NN O I-Generic
as NN O O
a NN O O
Quadratic NN O I-Task
Eigenvalue NN O I-Task
Problem NN O I-Task
offering NN O O
an NN O O
elegant NN O O
way NN O O
of NN O O
treating NN O O
nonlinear NN O I-OtherScientificTerm
monomials NN O I-OtherScientificTerm
and NN O O
delivering NN O O
a NN O O
quasi NN O I-Method
closed-form NN O I-Method
solution NN O I-Method
as NN O O
a NN O O
reliable NN O O
starting NN O O
point NN O O
for NN O O
a NN O O
further NN O O
bundle NN O I-Task
adjustment NN O I-Task
. NN O O

We NN O O
are NN O O
the NN O O
first NN O O
to NN O O
bring NN O O
the NN O O
closed NN O I-Method
form NN O I-Method
solution NN O I-Method
to NN O O
such NN O O
a NN O O
very NN O O
practical NN O O
problem NN O I-Generic
arising NN O O
in NN O O
video NN O I-Task
surveillance NN O I-Task
. NN O O

Results NN O O
in NN O O
different NN O O
camera NN O O
setups NN O O
demonstrate NN O O
the NN O O
feasibility NN O O
of NN O O
the NN O O
approach NN O I-Generic
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
human NN O I-Method
action NN O I-Method
recognition NN O I-Method
system NN O I-Method
suitable NN O O
for NN O O
embedded NN O I-Task
computer NN O I-Task
vision NN O I-Task
applications NN O I-Task
in NN O O
security NN O I-Task
systems NN O I-Task
, NN O O
human-computer NN O I-Task
interaction NN O I-Task
and NN O O
intelligent NN O I-Task
environments NN O I-Task
. NN O O

Our NN O O
system NN O I-Generic
is NN O O
suitable NN O O
for NN O O
embedded NN O I-Task
computer NN O I-Task
vision NN O I-Task
application NN O I-Task
based NN O O
on NN O O
three NN O O
reasons NN O O
. NN O O

Firstly NN O O
, NN O O
the NN O O
system NN O I-Generic
was NN O O
based NN O O
on NN O O
a NN O O
linear NN O I-Method
Support NN O I-Method
Vector NN O I-Method
Machine NN O I-Method
-LRB- NN O I-Method
SVM NN O I-Method
-RRB- NN O I-Method
classifier NN O I-Method
where NN O O
classification NN O I-Method
progress NN O I-Method
can NN O O
be NN O O
implemented NN O O
easily NN O O
and NN O O
quickly NN O O
in NN O O
embedded NN O I-OtherScientificTerm
hardware NN O I-OtherScientificTerm
. NN O O

Secondly NN O O
, NN O O
we NN O O
use NN O O
compacted NN O I-OtherScientificTerm
motion NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
easily NN O O
obtained NN O O
from NN O O
videos NN O I-Material
. NN O O

We NN O O
address NN O O
the NN O O
limitations NN O O
of NN O O
the NN O O
well NN O O
known NN O O
Motion NN O I-Method
History NN O I-Method
Image NN O I-Method
-LRB- NN O I-Method
MHI NN O I-Method
-RRB- NN O I-Method
and NN O O
propose NN O O
a NN O O
new NN O O
Hierarchical NN O I-Method
Motion NN O I-Method
History NN O I-Method
Histogram NN O I-Method
-LRB- NN O I-Method
HMHH NN O I-Method
-RRB- NN O I-Method
feature NN O I-Method
to NN O O
represent NN O O
the NN O O
motion NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
. NN O O

HMHH NN O I-Method
not NN O O
only NN O O
provides NN O O
rich NN O I-OtherScientificTerm
motion NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
, NN O O
but NN O O
also NN O O
remains NN O O
computationally NN O O
inexpensive NN O O
. NN O O

Finally NN O O
, NN O O
we NN O O
combine NN O O
MHI NN O I-Method
and NN O O
HMHH NN O I-Method
together NN O O
and NN O O
extract NN O O
a NN O O
low NN O I-OtherScientificTerm
dimension NN O I-OtherScientificTerm
feature NN O I-OtherScientificTerm
vector NN O I-OtherScientificTerm
to NN O O
be NN O O
used NN O O
in NN O O
the NN O O
SVM NN O I-Method
classifiers NN O I-Method
. NN O O

Experimental NN O O
results NN O O
show NN O O
that NN O O
our NN O O
system NN O I-Generic
achieves NN O O
significant NN O O
improvement NN O O
on NN O O
the NN O O
recognition NN O I-Task
performance NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
I NN O O
will NN O O
argue NN O O
for NN O O
a NN O O
model NN O I-Method
of NN O I-Method
grammatical NN O I-Method
processing NN O I-Method
that NN O O
is NN O O
based NN O O
on NN O O
uniform NN O I-Method
processing NN O I-Method
and NN O O
knowledge NN O I-Material
sources NN O I-Material
. NN O O

The NN O O
main NN O O
feature NN O O
of NN O O
this NN O O
model NN O I-Generic
is NN O O
to NN O O
view NN O O
parsing NN O I-Task
and NN O O
generation NN O I-Task
as NN O O
two NN O O
strongly NN O O
interleaved NN O O
tasks NN O I-Generic
performed NN O O
by NN O O
a NN O O
single NN O O
parametrized NN O I-Method
deduction NN O I-Method
process NN O I-Method
. NN O O

It NN O O
will NN O O
be NN O O
shown NN O O
that NN O O
this NN O O
view NN O O
supports NN O O
flexible NN O O
and NN O O
efficient NN O O
natural NN O I-Task
language NN O I-Task
processing NN O I-Task
. NN O O

-DOCSTART- NN O O

Link NN O I-Task
detection NN O I-Task
has NN O O
been NN O O
regarded NN O O
as NN O O
a NN O O
core NN O O
technology NN O O
for NN O O
the NN O O
Topic NN O I-Task
Detection NN O I-Task
and NN O I-Task
Tracking NN O I-Task
tasks NN O I-Task
of NN O I-Task
new NN O I-Task
event NN O I-Task
detection NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
formulate NN O O
story NN O I-Task
link NN O I-Task
detection NN O I-Task
and NN O O
new NN O I-Task
event NN O I-Task
detection NN O I-Task
as NN O O
information NN O I-Task
retrieval NN O I-Task
task NN O I-Task
and NN O O
hypothesize NN O O
on NN O O
the NN O O
impact NN O O
of NN O O
precision NN O I-Metric
and NN O O
recall NN O I-Metric
on NN O O
both NN O O
systems NN O I-Generic
. NN O O

Motivated NN O O
by NN O O
these NN O O
arguments NN O O
, NN O O
we NN O O
introduce NN O O
a NN O O
number NN O O
of NN O O
new NN O O
performance NN O I-Method
enhancing NN O I-Method
techniques NN O I-Method
including NN O O
part NN O I-Method
of NN O I-Method
speech NN O I-Method
tagging NN O I-Method
, NN O O
new NN O O
similarity NN O I-Method
measures NN O I-Method
and NN O O
expanded NN O I-OtherScientificTerm
stop NN O I-OtherScientificTerm
lists NN O I-OtherScientificTerm
. NN O O

Experimental NN O O
results NN O O
validate NN O O
our NN O O
hypothesis NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
attempt NN O O
to NN O O
understand NN O O
visual NN O I-Task
classification NN O I-Task
in NN O O
humans NN O O
using NN O O
both NN O O
psy-chophysical NN O I-Method
and NN O I-Method
machine NN O I-Method
learning NN O I-Method
techniques NN O I-Method
. NN O O

Frontal NN O I-OtherScientificTerm
views NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
human NN O I-OtherScientificTerm
faces NN O I-OtherScientificTerm
were NN O O
used NN O O
for NN O O
a NN O O
gender NN O I-Task
classification NN O I-Task
task NN O I-Task
. NN O O

Human NN O O
subjects NN O O
classified NN O O
the NN O O
faces NN O O
and NN O O
their NN O O
gender NN O I-OtherScientificTerm
judgment NN O I-OtherScientificTerm
, NN O O
reaction NN O I-OtherScientificTerm
time NN O I-OtherScientificTerm
and NN O O
confidence NN O I-OtherScientificTerm
rating NN O I-OtherScientificTerm
were NN O O
recorded NN O O
. NN O O

Several NN O O
hyperplane NN O I-Method
learning NN O I-Method
algorithms NN O I-Method
were NN O O
used NN O O
on NN O O
the NN O O
same NN O O
classification NN O I-Task
task NN O I-Task
using NN O O
the NN O O
Principal NN O I-OtherScientificTerm
Components NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
texture NN O I-OtherScientificTerm
and NN O O
flowfield NN O I-Method
representation NN O I-Method
of NN O I-Method
the NN O I-Method
faces NN O I-Method
. NN O O

The NN O O
classification NN O I-Task
performance NN O O
of NN O O
the NN O O
learning NN O I-Method
algorithms NN O I-Method
was NN O O
estimated NN O O
using NN O O
the NN O O
face NN O I-Task
database NN O I-Task
with NN O O
the NN O O
true NN O O
gender NN O O
of NN O O
the NN O O
faces NN O O
as NN O O
labels NN O O
, NN O O
and NN O O
also NN O O
with NN O O
the NN O O
gender NN O O
estimated NN O O
by NN O O
the NN O O
subjects NN O O
. NN O O

We NN O O
then NN O O
correlated NN O O
the NN O O
human NN O O
responses NN O O
to NN O O
the NN O O
distance NN O O
of NN O O
the NN O O
stimuli NN O O
to NN O O
the NN O O
separating NN O O
hyperplane NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
learning NN O I-Method
algorithms NN O I-Method
. NN O O

Our NN O O
results NN O O
suggest NN O O
that NN O O
human NN O I-Task
classification NN O I-Task
can NN O O
be NN O O
modeled NN O O
by NN O O
some NN O O
hyperplane NN O I-Method
algorithms NN O I-Method
in NN O O
the NN O O
feature NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
we NN O O
used NN O O
. NN O O

For NN O O
classification NN O I-Task
, NN O O
the NN O O
brain NN O O
needs NN O O
more NN O O
processing NN O O
for NN O O
stimuli NN O O
close NN O O
to NN O O
that NN O O
hyperplane NN O I-OtherScientificTerm
than NN O O
for NN O O
those NN O I-Generic
further NN O O
away NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
a NN O O
corpus-based NN O I-Method
supervised NN O I-Method
word NN O I-Method
sense NN O I-Method
disambiguation NN O I-Method
-LRB- NN O I-Method
WSD NN O I-Method
-RRB- NN O I-Method
system NN O I-Method
for NN O O
Dutch NN O I-Material
which NN O O
combines NN O O
statistical NN O I-Method
classification NN O I-Method
-LRB- NN O O
maximum NN O I-Method
entropy NN O I-Method
-RRB- NN O O
with NN O O
linguistic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
. NN O O

Instead NN O O
of NN O O
building NN O O
individual NN O O
classifiers NN O I-Method
per NN O O
ambiguous NN O I-OtherScientificTerm
wordform NN O I-OtherScientificTerm
, NN O O
we NN O O
introduce NN O O
a NN O O
lemma-based NN O I-Method
approach NN O I-Method
. NN O O

The NN O O
advantage NN O O
of NN O O
this NN O O
novel NN O O
method NN O I-Generic
is NN O O
that NN O O
it NN O I-Generic
clusters NN O O
all NN O O
inflected NN O I-OtherScientificTerm
forms NN O I-OtherScientificTerm
of NN O O
an NN O O
ambiguous NN O I-OtherScientificTerm
word NN O I-OtherScientificTerm
in NN O O
one NN O O
classifier NN O I-Method
, NN O O
therefore NN O O
augmenting NN O O
the NN O O
training NN O O
material NN O O
available NN O O
to NN O O
the NN O O
algorithm NN O I-Generic
. NN O O

Testing NN O O
the NN O O
lemma-based NN O I-Method
model NN O I-Method
on NN O O
the NN O O
Dutch NN O I-Material
Senseval-2 NN O I-Material
test NN O I-Material
data NN O I-Material
, NN O O
we NN O O
achieve NN O O
a NN O O
significant NN O O
increase NN O O
in NN O O
accuracy NN O O
over NN O O
the NN O O
wordform NN O I-Method
model NN O I-Method
. NN O O

Also NN O O
, NN O O
the NN O O
WSD NN O I-Method
system NN O I-Method
based NN O I-Method
on NN O I-Method
lemmas NN O I-Method
is NN O O
smaller NN O O
and NN O O
more NN O O
robust NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
an NN O O
exact NN O O
, NN O O
general NN O O
and NN O O
efficient NN O O
coarse-to-fine NN O I-Method
energy NN O I-Method
minimization NN O I-Method
strategy NN O I-Method
for NN O O
semantic NN O I-Task
video NN O I-Task
segmenta-tion NN O I-Task
. NN O O

Our NN O O
strategy NN O I-Generic
is NN O O
based NN O O
on NN O O
a NN O O
hierarchical NN O I-Task
abstraction NN O I-Task
of NN O I-Task
the NN O I-Task
supervoxel NN O I-Task
graph NN O I-Task
that NN O O
allows NN O O
us NN O O
to NN O O
minimize NN O O
an NN O O
energy NN O O
defined NN O O
at NN O O
the NN O O
finest NN O O
level NN O O
of NN O O
the NN O O
hierarchy NN O I-OtherScientificTerm
by NN O O
minimizing NN O O
a NN O O
series NN O O
of NN O O
simpler NN O O
energies NN O O
defined NN O O
over NN O O
coarser NN O I-OtherScientificTerm
graphs NN O I-OtherScientificTerm
. NN O O

The NN O O
strategy NN O I-Generic
is NN O O
exact NN O O
, NN O O
i.e. NN O O
, NN O O
it NN O I-Generic
produces NN O O
the NN O O
same NN O O
solution NN O O
as NN O O
minimizing NN O O
over NN O O
the NN O O
finest NN O I-OtherScientificTerm
graph NN O I-OtherScientificTerm
. NN O O

It NN O I-Generic
is NN O O
general NN O O
, NN O O
i.e. NN O O
, NN O O
it NN O I-Generic
can NN O O
be NN O O
used NN O O
to NN O O
minimize NN O O
any NN O O
energy NN O I-OtherScientificTerm
function NN O I-OtherScientificTerm
-LRB- NN O O
e.g. NN O O
, NN O O
unary NN O O
, NN O O
pairwise NN O O
, NN O O
and NN O O
higher-order NN O O
terms NN O O
-RRB- NN O O
with NN O O
any NN O O
existing NN O O
energy NN O I-Method
minimization NN O I-Method
algorithm NN O I-Method
-LRB- NN O O
e.g. NN O O
, NN O O
graph NN O I-Method
cuts NN O I-Method
and NN O O
belief NN O I-Method
propagation NN O I-Method
-RRB- NN O O
. NN O O

It NN O I-Generic
also NN O O
gives NN O O
significant NN O O
speedups NN O O
in NN O O
inference NN O I-Task
for NN O O
several NN O O
datasets NN O I-Generic
with NN O O
varying NN O O
degrees NN O O
of NN O O
spatio-temporal NN O I-OtherScientificTerm
continuity NN O I-OtherScientificTerm
. NN O O

We NN O O
also NN O O
discuss NN O O
the NN O O
strengths NN O O
and NN O O
weaknesses NN O O
of NN O O
our NN O O
strategy NN O I-Generic
relative NN O O
to NN O O
existing NN O O
hierarchical NN O I-Method
approaches NN O I-Method
, NN O O
and NN O O
the NN O O
kinds NN O O
of NN O O
image NN O I-Material
and NN O I-Material
video NN O I-Material
data NN O I-Material
that NN O O
provide NN O O
the NN O O
best NN O O
speedups NN O O
. NN O O

-DOCSTART- NN O O

Motivated NN O O
by NN O O
the NN O O
success NN O O
of NN O O
ensemble NN O I-Method
methods NN O I-Method
in NN O O
machine NN O I-Task
learning NN O I-Task
and NN O O
other NN O O
areas NN O O
of NN O O
natural NN O I-Task
language NN O I-Task
processing NN O I-Task
, NN O O
we NN O O
developed NN O O
a NN O O
multi-strategy NN O I-Method
and NN O I-Method
multi-source NN O I-Method
approach NN O I-Method
to NN O O
question NN O I-Task
answering NN O I-Task
which NN O O
is NN O O
based NN O O
on NN O O
combining NN O O
the NN O O
results NN O O
from NN O O
different NN O O
answering NN O I-Method
agents NN O I-Method
searching NN O O
for NN O O
answers NN O O
in NN O O
multiple NN O O
corpora NN O O
. NN O O

The NN O O
answering NN O I-Method
agents NN O I-Method
adopt NN O O
fundamentally NN O O
different NN O O
strategies NN O I-Generic
, NN O O
one NN O I-Generic
utilizing NN O O
primarily NN O O
knowledge-based NN O I-Method
mechanisms NN O I-Method
and NN O O
the NN O O
other NN O I-Generic
adopting NN O O
statistical NN O I-Method
techniques NN O I-Method
. NN O O

We NN O O
present NN O O
our NN O O
multi-level NN O I-Method
answer NN O I-Method
resolution NN O I-Method
algorithm NN O I-Method
that NN O O
combines NN O O
results NN O O
from NN O O
the NN O O
answering NN O I-Method
agents NN O I-Method
at NN O O
the NN O O
question NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
passage NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
and/or NN O I-OtherScientificTerm
answer NN O I-OtherScientificTerm
levels NN O I-OtherScientificTerm
. NN O O

Experiments NN O O
evaluating NN O O
the NN O O
effectiveness NN O O
of NN O O
our NN O O
answer NN O I-Method
resolution NN O I-Method
algorithm NN O I-Method
show NN O O
a NN O O
35.0 NN O O
% NN O O
relative NN O O
improvement NN O O
over NN O O
our NN O O
baseline NN O I-Generic
system NN O I-Generic
in NN O O
the NN O O
number NN O O
of NN O O
questions NN O O
correctly NN O O
answered NN O O
, NN O O
and NN O O
a NN O O
32.8 NN O O
% NN O O
improvement NN O O
according NN O O
to NN O O
the NN O O
average NN O I-Metric
precision NN O I-Metric
metric NN O I-Metric
. NN O O

-DOCSTART- NN O O

Word NN O I-Task
Identification NN O I-Task
has NN O O
been NN O O
an NN O O
important NN O O
and NN O O
active NN O O
issue NN O O
in NN O O
Chinese NN O I-Task
Natural NN O I-Task
Language NN O I-Task
Processing NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
a NN O O
new NN O O
mechanism NN O I-Generic
, NN O O
based NN O O
on NN O O
the NN O O
concept NN O O
of NN O O
sublanguage NN O I-OtherScientificTerm
, NN O O
is NN O O
proposed NN O O
for NN O O
identifying NN O O
unknown NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
, NN O O
especially NN O O
personal NN O I-OtherScientificTerm
names NN O I-OtherScientificTerm
, NN O O
in NN O O
Chinese NN O I-Material
newspapers NN O I-Material
. NN O O

The NN O O
proposed NN O O
mechanism NN O I-Generic
includes NN O O
title-driven NN O I-Task
name NN O I-Task
recognition NN O I-Task
, NN O O
adaptive NN O I-Task
dynamic NN O I-Task
word NN O I-Task
formation NN O I-Task
, NN O O
identification NN O I-Task
of NN O I-Task
2-character NN O I-Task
and NN O I-Task
3-character NN O I-Task
Chinese NN O I-Task
names NN O I-Task
without NN O I-Task
title NN O I-Task
. NN O O

We NN O O
will NN O O
show NN O O
the NN O O
experimental NN O O
results NN O O
for NN O O
two NN O O
corpora NN O O
and NN O O
compare NN O O
them NN O O
with NN O O
the NN O O
results NN O O
by NN O O
the NN O O
NTHU NN O I-Method
's NN O I-Method
statistic-based NN O I-Method
system NN O I-Method
, NN O O
the NN O O
only NN O O
system NN O I-Generic
that NN O O
we NN O O
know NN O O
has NN O O
attacked NN O O
the NN O O
same NN O O
problem NN O O
. NN O O

The NN O O
experimental NN O O
results NN O O
have NN O O
shown NN O O
significant NN O O
improvements NN O O
over NN O O
the NN O O
WI NN O I-Method
systems NN O I-Method
without NN O O
the NN O O
name NN O I-Task
identification NN O I-Task
capability NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
summarize NN O O
the NN O O
experiences NN O O
gained NN O O
from NN O O
a NN O O
field NN O O
trial NN O O
of NN O O
a NN O O
speaker NN O I-Method
verification NN O I-Method
system NN O I-Method
. NN O O

In NN O O
the NN O O
test NN O O
implementation NN O O
access NN O O
to NN O O
two NN O O
rooms NN O O
at NN O O
the NN O O
University NN O O
of NN O O
Frankfurt NN O O
had NN O O
been NN O O
controlled NN O O
by NN O O
a NN O O
speaker NN O I-Method
verification NN O I-Method
system NN O I-Method
. NN O O

The NN O O
paper NN O O
is NN O O
organized NN O O
as NN O O
follows NN O O
: NN O O
Firstly NN O O
, NN O O
we NN O O
will NN O O
describe NN O O
the NN O O
system NN O O
concepts NN O O
and NN O O
implementation NN O O
issues NN O O
. NN O O

Secondly NN O O
, NN O O
results NN O O
of NN O O
the NN O O
user NN O I-Method
evaluation NN O I-Method
are NN O O
reported NN O O
. NN O O

During NN O O
the NN O O
field NN O O
trial NN O O
all NN O O
speech NN O I-Material
data NN O I-Material
was NN O O
recorded NN O O
. NN O O

The NN O O
data NN O O
base NN O O
created NN O O
in NN O O
this NN O O
way NN O O
has NN O O
been NN O O
used NN O O
extensively NN O O
for NN O O
simulation NN O O
experiments NN O O
. NN O O

In NN O O
chapter NN O O
4 NN O O
we NN O O
will NN O O
describe NN O O
recent NN O O
experiments NN O O
focusing NN O O
on NN O O
the NN O O
use NN O O
of NN O O
Hidden NN O I-Method
Markov NN O I-Method
Models NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
report NN O O
describes NN O O
Paul NN O I-Method
, NN O O
a NN O O
computer NN O I-Method
text NN O I-Method
generation NN O I-Method
system NN O I-Method
designed NN O O
to NN O O
create NN O O
cohesive NN O I-Material
text NN O I-Material
through NN O O
the NN O O
use NN O O
of NN O O
lexical NN O I-OtherScientificTerm
substitutions NN O I-OtherScientificTerm
. NN O O

Specifically NN O O
, NN O O
this NN O O
system NN O I-Generic
is NN O O
designed NN O O
to NN O O
deterministically NN O O
choose NN O O
between NN O O
pronominalization NN O I-OtherScientificTerm
, NN O O
superordinate NN O I-OtherScientificTerm
substitution NN O I-OtherScientificTerm
, NN O O
and NN O O
definite NN O I-OtherScientificTerm
noun NN O I-OtherScientificTerm
phrase NN O I-OtherScientificTerm
reiteration NN O I-OtherScientificTerm
. NN O O

The NN O O
system NN O I-Generic
identifies NN O O
a NN O O
strength NN O O
of NN O O
antecedence NN O I-Task
recovery NN O I-Task
for NN O O
each NN O O
of NN O O
the NN O O
lexical NN O I-OtherScientificTerm
substitutions NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
addresses NN O O
the NN O O
problem NN O O
of NN O O
identifying NN O O
likely NN O O
topics NN O O
of NN O O
texts NN O O
by NN O O
their NN O O
position NN O O
in NN O O
the NN O O
text NN O O
. NN O O

It NN O O
describes NN O O
the NN O O
automated NN O O
training NN O O
and NN O O
evaluation NN O O
of NN O O
an NN O O
Optimal NN O I-Method
Position NN O I-Method
Policy NN O I-Method
, NN O O
a NN O O
method NN O I-Generic
of NN O O
locating NN O O
the NN O O
likely NN O O
positions NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
topic-bearing NN O I-OtherScientificTerm
sentences NN O I-OtherScientificTerm
based NN O O
on NN O O
genre-specific NN O I-OtherScientificTerm
regularities NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
discourse NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
. NN O O

This NN O O
method NN O I-Generic
can NN O O
be NN O O
used NN O O
in NN O O
applications NN O I-Generic
such NN O O
as NN O O
information NN O I-Task
retrieval NN O I-Task
, NN O O
routing NN O I-Task
, NN O O
and NN O O
text NN O I-Task
summarization NN O I-Task
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
a NN O O
general NN O O
framework NN O I-Generic
for NN O O
online NN O I-Task
multiclass NN O I-Task
learning NN O I-Task
based NN O O
on NN O O
the NN O O
notion NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
hypothesis NN O I-OtherScientificTerm
sharing NN O I-OtherScientificTerm
. NN O O

In NN O O
our NN O O
framework NN O I-Generic
sets NN O O
of NN O O
classes NN O O
are NN O O
associated NN O O
with NN O O
hypotheses NN O O
. NN O O

Thus NN O O
, NN O O
all NN O O
classes NN O O
within NN O O
a NN O O
given NN O O
set NN O O
share NN O O
the NN O O
same NN O O
hypothesis NN O O
. NN O O

This NN O O
framework NN O I-Generic
includes NN O O
as NN O O
special NN O O
cases NN O O
commonly NN O O
used NN O O
constructions NN O O
for NN O O
multiclass NN O I-Task
categorization NN O I-Task
such NN O O
as NN O O
allocating NN O O
a NN O O
unique NN O O
hypothesis NN O O
for NN O O
each NN O O
class NN O O
and NN O O
allocating NN O O
a NN O O
single NN O O
common NN O O
hypothesis NN O O
for NN O O
all NN O O
classes NN O O
. NN O O

We NN O O
generalize NN O O
the NN O O
multiclass NN O I-Method
Perceptron NN O I-Method
to NN O O
our NN O O
framework NN O I-Generic
and NN O O
derive NN O O
a NN O O
unifying NN O I-Method
mistake NN O I-Method
bound NN O I-Method
analysis NN O I-Method
. NN O O

Our NN O O
construction NN O O
naturally NN O O
extends NN O O
to NN O O
settings NN O O
where NN O O
the NN O O
number NN O O
of NN O O
classes NN O O
is NN O O
not NN O O
known NN O O
in NN O O
advance NN O O
but NN O O
, NN O O
rather NN O O
, NN O O
is NN O O
revealed NN O O
along NN O O
the NN O O
online NN O I-Method
learning NN O I-Method
process NN O I-Method
. NN O O

We NN O O
demonstrate NN O O
the NN O O
merits NN O O
of NN O O
our NN O O
approach NN O I-Generic
by NN O O
comparing NN O O
it NN O I-Generic
to NN O O
previous NN O O
methods NN O I-Generic
on NN O O
both NN O O
synthetic NN O I-Material
and NN O I-Material
natural NN O I-Material
datasets NN O I-Material
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
a NN O O
set NN O O
of NN O O
supervised NN O I-Method
machine NN O I-Method
learning NN O I-Method
experiments NN O O
centering NN O O
on NN O O
the NN O O
construction NN O O
of NN O O
statistical NN O I-Method
models NN O I-Method
of NN O I-Method
WH-questions NN O I-Method
. NN O O

These NN O O
models NN O I-Generic
, NN O O
which NN O O
are NN O O
built NN O O
from NN O O
shallow NN O I-OtherScientificTerm
linguistic NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
questions NN O I-OtherScientificTerm
, NN O O
are NN O O
employed NN O O
to NN O O
predict NN O O
target NN O O
variables NN O O
which NN O O
represent NN O O
a NN O O
user NN O I-OtherScientificTerm
's NN O I-OtherScientificTerm
informational NN O I-OtherScientificTerm
goals NN O I-OtherScientificTerm
. NN O O

We NN O O
report NN O O
on NN O O
different NN O O
aspects NN O O
of NN O O
the NN O O
predictive NN O O
performance NN O O
of NN O O
our NN O O
models NN O I-Generic
, NN O O
including NN O O
the NN O O
influence NN O O
of NN O O
various NN O O
training NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
testing NN O I-OtherScientificTerm
factors NN O I-OtherScientificTerm
on NN O O
predictive NN O O
performance NN O O
, NN O O
and NN O O
examine NN O O
the NN O O
relationships NN O O
among NN O O
the NN O O
target NN O O
variables NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
argue NN O O
in NN O O
favor NN O O
of NN O O
the NN O O
the NN O O
use NN O O
of NN O O
labeled NN O I-Method
directed NN O I-Method
graph NN O I-Method
to NN O O
represent NN O O
various NN O O
types NN O O
of NN O O
linguistic NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
, NN O O
and NN O O
illustrate NN O O
how NN O O
this NN O I-Generic
allows NN O O
one NN O O
to NN O O
view NN O O
NLP NN O I-Task
tasks NN O I-Task
as NN O O
graph NN O I-Task
transformations NN O I-Task
. NN O O

We NN O O
present NN O O
a NN O O
general NN O O
method NN O I-Method
for NN O O
learning NN O O
such NN O O
transformations NN O I-Generic
from NN O O
an NN O O
annotated NN O I-Material
corpus NN O I-Material
and NN O O
describe NN O O
experiments NN O O
with NN O O
two NN O O
applications NN O I-Generic
of NN O O
the NN O O
method NN O I-Generic
: NN O O
identification NN O I-Task
of NN O I-Task
non-local NN O I-Task
depenencies NN O I-Task
-LRB- NN O O
using NN O O
Penn NN O I-Material
Treebank NN O I-Material
data NN O I-Material
-RRB- NN O O
and NN O O
semantic NN O I-Task
role NN O I-Task
labeling NN O I-Task
-LRB- NN O O
using NN O O
Proposition NN O I-Material
Bank NN O I-Material
data NN O I-Material
-RRB- NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
describe NN O O
a NN O O
generative NN O I-Method
probabilistic NN O I-Method
model NN O I-Method
of NN O I-Method
natural NN O I-Method
language NN O I-Method
, NN O O
which NN O O
we NN O O
call NN O O
HBG NN O I-Method
, NN O O
that NN O O
takes NN O O
advantage NN O O
of NN O O
detailed NN O O
linguistic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
to NN O O
resolve NN O O
ambiguity NN O I-Task
. NN O O

HBG NN O I-Method
incorporates NN O O
lexical NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
syntactic NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
semantic NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
structural NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
from NN O O
the NN O O
parse NN O I-OtherScientificTerm
tree NN O I-OtherScientificTerm
into NN O O
the NN O O
disambiguation NN O I-Task
process NN O I-Task
in NN O O
a NN O O
novel NN O O
way NN O O
. NN O O

We NN O O
use NN O O
a NN O O
corpus NN O I-Material
of NN O I-Material
bracketed NN O I-Material
sentences NN O I-Material
, NN O O
called NN O O
a NN O O
Treebank NN O I-Material
, NN O O
in NN O O
combination NN O O
with NN O O
decision NN O I-Method
tree NN O I-Method
building NN O I-Method
to NN O O
tease NN O O
out NN O O
the NN O O
relevant NN O O
aspects NN O O
of NN O O
a NN O O
parse NN O I-OtherScientificTerm
tree NN O I-OtherScientificTerm
that NN O O
will NN O O
determine NN O O
the NN O O
correct NN O O
parse NN O I-OtherScientificTerm
of NN O O
a NN O O
sentence NN O O
. NN O O

This NN O O
stands NN O O
in NN O O
contrast NN O O
to NN O O
the NN O O
usual NN O O
approach NN O O
of NN O O
further NN O O
grammar NN O I-Task
tailoring NN O I-Task
via NN O O
the NN O O
usual NN O O
linguistic NN O I-Method
introspection NN O I-Method
in NN O O
the NN O O
hope NN O O
of NN O O
generating NN O O
the NN O O
correct NN O O
parse NN O I-OtherScientificTerm
. NN O O

In NN O O
head-to-head NN O I-OtherScientificTerm
tests NN O I-OtherScientificTerm
against NN O O
one NN O O
of NN O O
the NN O O
best NN O O
existing NN O O
robust NN O I-Method
probabilistic NN O I-Method
parsing NN O I-Method
models NN O I-Method
, NN O O
which NN O O
we NN O O
call NN O O
P-CFG NN O I-Method
, NN O O
the NN O O
HBG NN O I-Method
model NN O I-Method
significantly NN O O
outperforms NN O O
P-CFG NN O I-Method
, NN O O
increasing NN O O
the NN O O
parsing NN O I-Metric
accuracy NN O I-Metric
rate NN O I-Metric
from NN O O
60 NN O O
% NN O O
to NN O O
75 NN O O
% NN O O
, NN O O
a NN O O
37 NN O O
% NN O O
reduction NN O O
in NN O O
error NN O O
. NN O O

-DOCSTART- NN O O

The NN O O
verb NN O I-OtherScientificTerm
forms NN O I-OtherScientificTerm
are NN O O
often NN O O
claimed NN O O
to NN O O
convey NN O O
two NN O O
kinds NN O O
of NN O O
information NN O I-Generic
: NN O O
1 NN O O
. NN O O

whether NN O O
the NN O O
event NN O O
described NN O O
in NN O O
a NN O O
sentence NN O O
is NN O O
present NN O O
, NN O O
past NN O O
or NN O O
future NN O O
-LRB- NN O O
= NN O O
deictic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
-RRB- NN O O
2 NN O O
. NN O O

whether NN O O
the NN O O
event NN O O
described NN O O
in NN O O
a NN O O
sentence NN O O
is NN O O
presented NN O O
as NN O O
completed NN O O
, NN O O
going NN O O
on NN O O
, NN O O
just NN O O
starting NN O O
or NN O O
being NN O O
finished NN O O
-LRB- NN O O
= NN O O
aspectual NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
-RRB- NN O O
. NN O O

It NN O O
will NN O O
be NN O O
demonstrated NN O O
in NN O O
this NN O O
paper NN O O
that NN O O
one NN O O
has NN O O
to NN O O
add NN O O
a NN O O
third NN O O
component NN O O
to NN O O
the NN O O
analysis NN O I-Task
of NN O I-Task
verb NN O I-Task
form NN O I-Task
meanings NN O I-Task
, NN O O
namely NN O O
whether NN O O
or NN O O
not NN O O
they NN O O
express NN O O
habituality NN O O
. NN O O

The NN O O
framework NN O O
of NN O O
the NN O O
analysis NN O I-Generic
is NN O O
model-theoretic NN O I-OtherScientificTerm
semantics NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
addresses NN O O
the NN O O
issue NN O O
of NN O O
word-sense NN O I-Task
ambiguity NN O I-Task
in NN O O
extraction NN O O
from NN O O
machine-readable NN O I-Material
resources NN O I-Material
for NN O O
the NN O O
construction NN O I-Task
of NN O I-Task
large-scale NN O I-Task
knowledge NN O I-Task
sources NN O I-Task
. NN O O

We NN O O
describe NN O O
two NN O O
experiments NN O O
: NN O O
one NN O O
which NN O O
ignored NN O O
word-sense NN O I-OtherScientificTerm
distinctions NN O I-OtherScientificTerm
, NN O O
resulting NN O O
in NN O O
6.3 NN O O
% NN O O
accuracy NN O I-Metric
for NN O O
semantic NN O I-Method
classification NN O I-Method
of NN O O
verbs NN O O
based NN O O
on NN O O
-LRB- NN O O
Levin NN O O
, NN O O
1993 NN O O
-RRB- NN O O
; NN O O
and NN O O
one NN O O
which NN O O
exploited NN O O
word-sense NN O I-OtherScientificTerm
distinctions NN O I-OtherScientificTerm
, NN O O
resulting NN O O
in NN O O
97.9 NN O O
% NN O O
accuracy NN O I-Metric
. NN O O

These NN O O
experiments NN O O
were NN O O
dual NN O O
purpose NN O O
: NN O O
-LRB- NN O O
1 NN O O
-RRB- NN O O
to NN O O
validate NN O O
the NN O O
central NN O O
thesis NN O O
of NN O O
the NN O O
work NN O O
of NN O O
-LRB- NN O O
Levin NN O O
, NN O O
1993 NN O O
-RRB- NN O O
, NN O O
i.e. NN O O
, NN O O
that NN O O
verb NN O I-OtherScientificTerm
semantics NN O I-OtherScientificTerm
and NN O O
syntactic NN O I-OtherScientificTerm
behavior NN O I-OtherScientificTerm
are NN O O
predictably NN O O
related NN O O
; NN O O
-LRB- NN O O
2 NN O O
-RRB- NN O O
to NN O O
demonstrate NN O O
that NN O O
a NN O O
15-fold NN O O
improvement NN O O
can NN O O
be NN O O
achieved NN O O
in NN O O
deriving NN O O
semantic NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
from NN O O
syntactic NN O I-OtherScientificTerm
cues NN O I-OtherScientificTerm
if NN O O
we NN O O
first NN O O
divide NN O O
the NN O O
syntactic NN O I-OtherScientificTerm
cues NN O I-OtherScientificTerm
into NN O O
distinct NN O O
groupings NN O O
that NN O O
correlate NN O O
with NN O O
different NN O O
word NN O I-OtherScientificTerm
senses NN O I-OtherScientificTerm
. NN O O

Finally NN O O
, NN O O
we NN O O
show NN O O
that NN O O
we NN O O
can NN O O
provide NN O O
effective NN O O
acquisition NN O O
techniques NN O I-Generic
for NN O O
novel NN O O
word NN O I-OtherScientificTerm
senses NN O I-OtherScientificTerm
using NN O O
a NN O O
combination NN O O
of NN O O
online NN O I-Material
sources NN O I-Material
. NN O O

-DOCSTART- NN O O

The NN O O
TIPSTER NN O I-Method
Architecture NN O I-Method
has NN O O
been NN O O
designed NN O O
to NN O O
enable NN O O
a NN O O
variety NN O O
of NN O O
different NN O O
text NN O I-Task
applications NN O I-Task
to NN O O
use NN O O
a NN O O
set NN O O
of NN O O
common NN O I-Method
text NN O I-Method
processing NN O I-Method
modules NN O I-Method
. NN O O

Since NN O O
user NN O I-OtherScientificTerm
interfaces NN O I-OtherScientificTerm
work NN O O
best NN O O
when NN O O
customized NN O O
for NN O O
particular NN O O
applications NN O I-Generic
, NN O O
it NN O O
is NN O O
appropriator NN O O
that NN O O
no NN O O
particular NN O O
user NN O I-OtherScientificTerm
interface NN O I-OtherScientificTerm
styles NN O I-OtherScientificTerm
or NN O I-OtherScientificTerm
conventions NN O I-OtherScientificTerm
are NN O O
described NN O O
in NN O O
the NN O O
TIPSTER NN O I-OtherScientificTerm
Architecture NN O I-OtherScientificTerm
specification NN O I-OtherScientificTerm
. NN O O

However NN O O
, NN O O
the NN O O
Computing NN O O
Research NN O O
Laboratory NN O O
-LRB- NN O O
CRL NN O O
-RRB- NN O O
has NN O O
constructed NN O O
several NN O O
TIPSTER NN O I-Task
applications NN O I-Task
that NN O O
use NN O O
a NN O O
common NN O O
set NN O O
of NN O O
configurable NN O O
Graphical NN O I-OtherScientificTerm
User NN O I-OtherScientificTerm
Interface NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
GUI NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
functions NN O I-OtherScientificTerm
. NN O O

These NN O O
GUIs NN O I-OtherScientificTerm
were NN O O
constructed NN O O
using NN O O
CRL NN O I-Method
's NN O I-Method
TIPSTER NN O I-Method
User NN O I-Method
Interface NN O I-Method
Toolkit NN O I-Method
-LRB- NN O I-Method
TUIT NN O I-Method
-RRB- NN O I-Method
. NN O O

TUIT NN O I-Method
is NN O O
a NN O O
software NN O I-Method
library NN O I-Method
that NN O O
can NN O O
be NN O O
used NN O O
to NN O O
construct NN O O
multilingual NN O I-OtherScientificTerm
TIPSTER NN O I-OtherScientificTerm
user NN O I-OtherScientificTerm
interfaces NN O I-OtherScientificTerm
for NN O O
a NN O O
set NN O O
of NN O O
common NN O O
user NN O O
tasks NN O O
. NN O O

CRL NN O O
developed NN O O
TUIT NN O I-Method
to NN O O
support NN O O
their NN O O
work NN O O
to NN O O
integrate NN O O
TIPSTER NN O I-Method
modules NN O I-Method
for NN O O
the NN O O
6 NN O O
and NN O O
12 NN O O
month NN O O
TIPSTER NN O O
II NN O O
demonstrations NN O O
as NN O O
well NN O O
as NN O O
their NN O O
Oleada NN O O
and NN O O
Temple NN O O
demonstration NN O O
projects NN O O
. NN O O

This NN O O
paper NN O O
briefly NN O O
describes NN O O
TUIT NN O I-Method
and NN O O
its NN O O
capabilities NN O O
. NN O O

-DOCSTART- NN O O

Current NN O O
natural NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
interfaces NN O I-OtherScientificTerm
have NN O O
concentrated NN O O
largely NN O O
on NN O O
determining NN O O
the NN O O
literal NN O O
meaning NN O O
of NN O O
input NN O O
from NN O O
their NN O O
users NN O O
. NN O O

While NN O O
such NN O O
decoding NN O I-Method
is NN O O
an NN O O
essential NN O O
underpinning NN O O
, NN O O
much NN O O
recent NN O O
work NN O O
suggests NN O O
that NN O O
natural NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
interfaces NN O I-OtherScientificTerm
will NN O O
never NN O O
appear NN O O
cooperative NN O O
or NN O O
graceful NN O O
unless NN O O
they NN O I-Generic
also NN O O
incorporate NN O O
numerous NN O O
non-literal NN O I-Method
aspects NN O I-Method
of NN O I-Method
communication NN O I-Method
, NN O O
such NN O O
as NN O O
robust NN O I-Method
communication NN O I-Method
procedures NN O I-Method
. NN O O

This NN O O
paper NN O O
defends NN O O
that NN O O
view NN O O
, NN O O
but NN O O
claims NN O O
that NN O O
direct NN O O
imitation NN O O
of NN O O
human NN O O
performance NN O O
is NN O O
not NN O O
the NN O O
best NN O O
way NN O O
to NN O O
implement NN O O
many NN O O
of NN O O
these NN O O
non-literal NN O I-Method
aspects NN O I-Method
of NN O I-Method
communication NN O I-Method
; NN O O
that NN O O
the NN O O
new NN O O
technology NN O O
of NN O O
powerful NN O O
personal NN O I-OtherScientificTerm
computers NN O I-OtherScientificTerm
with NN O O
integral NN O O
graphics NN O I-OtherScientificTerm
displays NN O I-OtherScientificTerm
offers NN O O
techniques NN O O
superior NN O O
to NN O O
those NN O O
of NN O O
humans NN O O
for NN O O
these NN O O
aspects NN O O
, NN O O
while NN O O
still NN O O
satisfying NN O O
human NN O O
communication NN O O
needs NN O O
. NN O O

The NN O O
paper NN O O
proposes NN O O
interfaces NN O O
based NN O O
on NN O O
a NN O O
judicious NN O O
mixture NN O O
of NN O O
these NN O O
techniques NN O O
and NN O O
the NN O O
still NN O O
valuable NN O O
methods NN O O
of NN O O
more NN O O
traditional NN O O
natural NN O I-OtherScientificTerm
language NN O I-OtherScientificTerm
interfaces NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
a NN O O
framework NN O O
in NN O O
which NN O O
Lagrangian NN O I-Method
Particle NN O I-Method
Dynamics NN O I-Method
is NN O O
used NN O O
for NN O O
the NN O O
segmentation NN O I-Task
of NN O I-Task
high NN O I-Task
density NN O I-Task
crowd NN O I-Task
flows NN O I-Task
and NN O O
detection NN O I-Task
of NN O I-Task
flow NN O I-Task
instabilities NN O I-Task
. NN O O

For NN O O
this NN O O
purpose NN O O
, NN O O
a NN O O
flow NN O I-OtherScientificTerm
field NN O I-OtherScientificTerm
generated NN O O
by NN O O
a NN O O
moving NN O I-OtherScientificTerm
crowd NN O I-OtherScientificTerm
is NN O O
treated NN O O
as NN O O
an NN O O
aperiodic NN O I-Method
dynamical NN O I-Method
system NN O I-Method
. NN O O

A NN O O
grid NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
particles NN O I-OtherScientificTerm
is NN O O
overlaid NN O O
on NN O O
the NN O O
flow NN O I-OtherScientificTerm
field NN O I-OtherScientificTerm
, NN O O
and NN O O
is NN O O
advected NN O O
using NN O O
a NN O O
numerical NN O I-Method
integration NN O I-Method
scheme NN O I-Method
. NN O O

The NN O O
evolution NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
particles NN O I-OtherScientificTerm
through NN O O
the NN O O
flow NN O O
is NN O O
tracked NN O O
using NN O O
a NN O O
Flow NN O I-Method
Map NN O I-Method
, NN O O
whose NN O O
spatial NN O I-OtherScientificTerm
gradients NN O I-OtherScientificTerm
are NN O O
subsequently NN O O
used NN O O
to NN O O
setup NN O O
a NN O O
Cauchy NN O I-Method
Green NN O I-Method
Deformation NN O I-Method
tensor NN O I-Method
for NN O O
quantifying NN O O
the NN O O
amount NN O O
by NN O O
which NN O O
the NN O O
neighboring NN O O
particles NN O O
have NN O O
diverged NN O O
over NN O O
the NN O O
length NN O O
of NN O O
the NN O O
integration NN O O
. NN O O

The NN O O
maximum NN O I-OtherScientificTerm
eigenvalue NN O I-OtherScientificTerm
of NN O O
the NN O O
tensor NN O I-Generic
is NN O O
used NN O O
to NN O O
construct NN O O
a NN O O
Finite NN O I-OtherScientificTerm
Time NN O I-OtherScientificTerm
Lyapunov NN O I-OtherScientificTerm
Exponent NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
FTLE NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
field NN O I-OtherScientificTerm
, NN O O
which NN O O
reveals NN O O
the NN O O
Lagrangian NN O I-OtherScientificTerm
Coherent NN O I-OtherScientificTerm
Structures NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
LCS NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
present NN O O
in NN O O
the NN O O
underlying NN O O
flow NN O O
. NN O O

The NN O O
LCS NN O I-OtherScientificTerm
divide NN O O
flow NN O O
into NN O O
regions NN O O
of NN O O
qualitatively NN O O
different NN O O
dynamics NN O O
and NN O O
are NN O O
used NN O O
to NN O O
locate NN O O
boundaries NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
flow NN O I-OtherScientificTerm
segments NN O I-OtherScientificTerm
in NN O O
a NN O O
normalized NN O I-Method
cuts NN O I-Method
framework NN O I-Method
. NN O O

Any NN O O
change NN O O
in NN O O
the NN O O
number NN O O
of NN O O
flow NN O O
segments NN O O
over NN O O
time NN O O
is NN O O
regarded NN O O
as NN O O
an NN O O
instability NN O O
, NN O O
which NN O O
is NN O O
detected NN O O
by NN O O
establishing NN O O
correspondences NN O O
between NN O O
flow NN O O
segments NN O O
over NN O O
time NN O O
. NN O O

The NN O O
experiments NN O O
are NN O O
conducted NN O O
on NN O O
a NN O O
challenging NN O O
set NN O O
of NN O O
videos NN O O
taken NN O O
from NN O O
Google NN O I-Material
Video NN O I-Material
and NN O O
a NN O O
National NN O I-Material
Geographic NN O I-Material
documentary NN O I-Material
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
study NN O O
a NN O O
set NN O O
of NN O O
problems NN O O
that NN O O
are NN O O
of NN O O
considerable NN O O
importance NN O O
to NN O O
Statistical NN O I-Task
Machine NN O I-Task
Translation NN O I-Task
-LRB- NN O I-Task
SMT NN O I-Task
-RRB- NN O I-Task
but NN O O
which NN O O
have NN O O
not NN O O
been NN O O
addressed NN O O
satisfactorily NN O O
by NN O O
the NN O O
SMT NN O O
research NN O O
community NN O O
. NN O O

Over NN O O
the NN O O
last NN O O
decade NN O O
, NN O O
a NN O O
variety NN O O
of NN O O
SMT NN O I-Method
algorithms NN O I-Method
have NN O O
been NN O O
built NN O O
and NN O O
empirically NN O O
tested NN O O
whereas NN O O
little NN O O
is NN O O
known NN O O
about NN O O
the NN O O
computational NN O I-Metric
complexity NN O I-Metric
of NN O O
some NN O O
of NN O O
the NN O O
fundamental NN O O
problems NN O I-Generic
of NN O O
SMT NN O I-Task
. NN O O

Our NN O O
work NN O O
aims NN O O
at NN O O
providing NN O O
useful NN O O
insights NN O O
into NN O O
the NN O O
the NN O O
computational NN O I-Metric
complexity NN O I-Metric
of NN O O
those NN O O
problems NN O I-Generic
. NN O O

We NN O O
prove NN O O
that NN O O
while NN O O
IBM NN O I-Method
Models NN O I-Method
1-2 NN O I-Method
are NN O O
conceptually NN O O
and NN O O
computationally NN O O
simple NN O O
, NN O O
computations NN O I-Generic
involving NN O O
the NN O O
higher NN O O
-LRB- NN O O
and NN O O
more NN O O
useful NN O O
-RRB- NN O O
models NN O I-Generic
are NN O O
hard NN O O
. NN O O

Since NN O O
it NN O O
is NN O O
unlikely NN O O
that NN O O
there NN O O
exists NN O O
a NN O O
polynomial NN O I-Method
time NN O I-Method
solution NN O I-Method
for NN O O
any NN O O
of NN O O
these NN O O
hard NN O I-Generic
problems NN O I-Generic
-LRB- NN O O
unless NN O O
P NN O O
= NN O O
NP NN O O
and NN O O
P NN O O
#P NN O O
= NN O O
P NN O O
-RRB- NN O O
, NN O O
our NN O O
results NN O O
highlight NN O O
and NN O O
justify NN O O
the NN O O
need NN O O
for NN O O
developing NN O O
polynomial NN O I-Method
time NN O I-Method
approximations NN O I-Method
for NN O O
these NN O O
computations NN O I-Generic
. NN O O

We NN O O
also NN O O
discuss NN O O
some NN O O
practical NN O O
ways NN O O
of NN O O
dealing NN O O
with NN O O
complexity NN O I-Metric
. NN O O

-DOCSTART- NN O O

Most NN O O
state-of-the-art NN O O
evaluation NN O I-Metric
measures NN O I-Metric
for NN O O
machine NN O I-Task
translation NN O I-Task
assign NN O O
high NN O O
costs NN O O
to NN O O
movements NN O O
of NN O O
word NN O O
blocks NN O O
. NN O O

In NN O O
many NN O O
cases NN O O
though NN O O
such NN O O
movements NN O O
still NN O O
result NN O O
in NN O O
correct NN O O
or NN O O
almost NN O O
correct NN O O
sentences NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
will NN O O
present NN O O
a NN O O
new NN O O
evaluation NN O I-Metric
measure NN O I-Metric
which NN O O
explicitly NN O O
models NN O O
block NN O I-OtherScientificTerm
reordering NN O I-OtherScientificTerm
as NN O O
an NN O O
edit NN O I-OtherScientificTerm
operation NN O I-OtherScientificTerm
. NN O O

Our NN O O
measure NN O I-Generic
can NN O O
be NN O O
exactly NN O O
calculated NN O O
in NN O O
quadratic NN O I-OtherScientificTerm
time NN O I-OtherScientificTerm
. NN O O

Furthermore NN O O
, NN O O
we NN O O
will NN O O
show NN O O
how NN O O
some NN O O
evaluation NN O I-Metric
measures NN O I-Metric
can NN O O
be NN O O
improved NN O O
by NN O O
the NN O O
introduction NN O O
of NN O O
word-dependent NN O I-OtherScientificTerm
substitution NN O I-OtherScientificTerm
costs NN O I-OtherScientificTerm
. NN O O

The NN O O
correlation NN O O
of NN O O
the NN O O
new NN O O
measure NN O I-Generic
with NN O O
human NN O I-OtherScientificTerm
judgment NN O I-OtherScientificTerm
has NN O O
been NN O O
investigated NN O O
systematically NN O O
on NN O O
two NN O O
different NN O O
language NN O I-Material
pairs NN O I-Material
. NN O O

The NN O O
experimental NN O O
results NN O O
will NN O O
show NN O O
that NN O O
it NN O I-Generic
significantly NN O O
outperforms NN O O
state-of-the-art NN O O
approaches NN O I-Generic
in NN O O
sentence-level NN O I-Metric
correlation NN O I-Metric
. NN O O

Results NN O O
from NN O O
experiments NN O O
with NN O O
word NN O I-OtherScientificTerm
dependent NN O I-OtherScientificTerm
substitution NN O I-OtherScientificTerm
costs NN O I-OtherScientificTerm
will NN O O
demonstrate NN O O
an NN O O
additional NN O O
increase NN O O
of NN O O
correlation NN O O
between NN O O
automatic NN O I-Metric
evaluation NN O I-Metric
measures NN O I-Metric
and NN O O
human NN O I-OtherScientificTerm
judgment NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

The NN O O
Rete NN O I-Method
and NN O I-Method
Treat NN O I-Method
algorithms NN O I-Method
are NN O O
considered NN O O
the NN O O
most NN O O
efficient NN O O
implementation NN O I-Method
techniques NN O I-Method
for NN O O
Forward NN O I-Task
Chaining NN O I-Task
rule NN O I-Task
systems NN O I-Task
. NN O O

These NN O O
algorithms NN O I-Generic
support NN O O
a NN O O
language NN O I-Method
of NN O I-Method
limited NN O I-Method
expressive NN O I-Method
power NN O I-Method
. NN O O

Assertions NN O I-OtherScientificTerm
are NN O O
not NN O O
allowed NN O O
to NN O O
contain NN O O
variables NN O I-OtherScientificTerm
, NN O O
making NN O O
universal NN O I-OtherScientificTerm
quantification NN O I-OtherScientificTerm
impossible NN O O
to NN O O
express NN O O
except NN O O
as NN O O
a NN O O
rule NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
show NN O O
how NN O O
to NN O O
support NN O O
full NN O I-Task
unification NN O I-Task
in NN O O
these NN O O
algorithms NN O I-Generic
. NN O O

We NN O O
also NN O O
show NN O O
that NN O O
: NN O O
Supporting NN O O
full NN O I-Task
unification NN O I-Task
is NN O O
costly NN O O
; NN O O
Full NN O I-Task
unification NN O I-Task
is NN O O
not NN O O
used NN O O
frequently NN O O
; NN O O
A NN O O
combination NN O O
of NN O O
compile NN O I-Metric
time NN O I-Metric
and NN O O
run NN O I-Metric
time NN O I-Metric
checks NN O O
can NN O O
determine NN O O
when NN O O
full NN O I-Task
unification NN O I-Task
is NN O O
not NN O O
needed NN O O
. NN O O

We NN O O
present NN O O
data NN O O
to NN O O
show NN O O
that NN O O
the NN O O
cost NN O O
of NN O O
supporting NN O O
full NN O I-Task
unification NN O I-Task
can NN O O
be NN O O
reduced NN O O
in NN O O
proportion NN O O
to NN O O
the NN O O
degree NN O O
that NN O O
it NN O I-Generic
is NN O O
n't NN O O
employed NN O O
and NN O O
that NN O O
for NN O O
many NN O O
practical NN O O
systems NN O O
this NN O O
cost NN O O
is NN O O
negligible NN O O
. NN O O

-DOCSTART- NN O O

A NN O O
method NN O I-Generic
for NN O O
error NN O I-Task
correction NN O I-Task
of NN O O
ill-formed NN O I-Material
input NN O I-Material
is NN O O
described NN O O
that NN O O
acquires NN O O
dialogue NN O I-OtherScientificTerm
patterns NN O I-OtherScientificTerm
in NN O O
typical NN O O
usage NN O O
and NN O O
uses NN O O
these NN O O
patterns NN O I-Generic
to NN O O
predict NN O O
new NN O O
inputs NN O O
. NN O O

Error NN O I-Task
correction NN O I-Task
is NN O O
done NN O O
by NN O O
strongly NN O O
biasing NN O O
parsing NN O I-Task
toward NN O O
expected NN O O
meanings NN O O
unless NN O O
clear NN O O
evidence NN O O
from NN O O
the NN O O
input NN O O
shows NN O O
the NN O O
current NN O O
sentence NN O O
is NN O O
not NN O O
expected NN O O
. NN O O

A NN O O
dialogue NN O I-Method
acquisition NN O I-Method
and NN O I-Method
tracking NN O I-Method
algorithm NN O I-Method
is NN O O
presented NN O O
along NN O O
with NN O O
a NN O O
description NN O O
of NN O O
its NN O O
implementation NN O O
in NN O O
a NN O O
voice NN O I-Method
interactive NN O I-Method
system NN O I-Method
. NN O O

A NN O O
series NN O O
of NN O O
tests NN O O
are NN O O
described NN O O
that NN O O
show NN O O
the NN O O
power NN O O
of NN O O
the NN O O
error NN O I-Method
correction NN O I-Method
methodology NN O I-Method
when NN O O
stereotypic NN O I-Material
dialogue NN O I-Material
occurs NN O O
. NN O O

-DOCSTART- NN O O

Traditional NN O O
linear NN O I-Method
Fukunaga-Koontz NN O I-Method
Transform NN O I-Method
-LRB- NN O I-Method
FKT NN O I-Method
-RRB- NN O I-Method
-LSB- NN O O
1 NN O O
-RSB- NN O O
is NN O O
a NN O O
powerful NN O O
discriminative NN O I-Method
subspaces NN O I-Method
building NN O I-Method
approach NN O I-Method
. NN O O

Previous NN O O
work NN O O
has NN O O
successfully NN O O
extended NN O O
FKT NN O I-Method
to NN O O
be NN O O
able NN O O
to NN O O
deal NN O O
with NN O O
small-sample-size NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
extend NN O O
traditional NN O O
linear NN O I-Method
FKT NN O I-Method
to NN O O
enable NN O O
it NN O I-Generic
to NN O O
work NN O O
in NN O O
multi-class NN O I-Task
problem NN O I-Task
and NN O O
also NN O O
in NN O O
higher NN O I-OtherScientificTerm
dimensional NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
kernel NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
subspaces NN O I-OtherScientificTerm
and NN O O
therefore NN O O
provide NN O O
enhanced NN O O
discrimination NN O I-OtherScientificTerm
ability NN O I-OtherScientificTerm
. NN O O

We NN O O
verify NN O O
the NN O O
effectiveness NN O O
of NN O O
the NN O O
proposed NN O O
Kernel NN O I-Method
Fukunaga-Koontz NN O I-Method
Transform NN O I-Method
by NN O O
demonstrating NN O O
its NN O O
effectiveness NN O O
in NN O O
face NN O I-Task
recognition NN O I-Task
applications NN O I-Task
; NN O O
however NN O O
the NN O O
proposed NN O O
non-linear NN O I-Method
generalization NN O I-Method
can NN O O
be NN O O
applied NN O O
to NN O O
any NN O O
other NN O O
domain NN O I-Task
specific NN O I-Task
problems NN O I-Task
. NN O O

-DOCSTART- NN O O

The NN O O
study NN O O
addresses NN O O
the NN O O
problem NN O O
of NN O O
automatic NN O I-Task
acquisition NN O I-Task
of NN O I-Task
entailment NN O I-Task
relations NN O I-Task
between NN O O
verbs NN O O
. NN O O

While NN O O
this NN O O
task NN O I-Generic
has NN O O
much NN O O
in NN O O
common NN O O
with NN O O
paraphrases NN O I-Task
acquisition NN O I-Task
which NN O O
aims NN O O
to NN O O
discover NN O O
semantic NN O I-OtherScientificTerm
equivalence NN O I-OtherScientificTerm
between NN O O
verbs NN O O
, NN O O
the NN O O
main NN O O
challenge NN O O
of NN O O
entailment NN O I-Task
acquisition NN O I-Task
is NN O O
to NN O O
capture NN O O
asymmetric NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
or NN O I-OtherScientificTerm
directional NN O I-OtherScientificTerm
, NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
. NN O O

Motivated NN O O
by NN O O
the NN O O
intuition NN O O
that NN O O
it NN O O
often NN O O
underlies NN O O
the NN O O
local NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
coherent NN O I-OtherScientificTerm
text NN O I-OtherScientificTerm
, NN O O
we NN O O
develop NN O O
a NN O O
method NN O I-Generic
that NN O O
discovers NN O O
verb NN O I-Task
entailment NN O I-Task
using NN O O
evidence NN O O
about NN O O
discourse NN O I-OtherScientificTerm
relations NN O I-OtherScientificTerm
between NN O O
clauses NN O O
available NN O O
in NN O O
a NN O O
parsed NN O I-Material
corpus NN O I-Material
. NN O O

In NN O O
comparison NN O O
with NN O O
earlier NN O O
work NN O O
, NN O O
the NN O O
proposed NN O O
method NN O I-Generic
covers NN O O
a NN O O
much NN O O
wider NN O O
range NN O O
of NN O O
verb NN O I-OtherScientificTerm
entailment NN O I-OtherScientificTerm
types NN O I-OtherScientificTerm
and NN O O
learns NN O O
the NN O O
mapping NN O I-Task
between NN O I-Task
verbs NN O I-Task
with NN O O
highly NN O I-OtherScientificTerm
varied NN O I-OtherScientificTerm
argument NN O I-OtherScientificTerm
structures NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
cast NN O O
the NN O O
problem NN O O
of NN O O
point NN O I-Task
cloud NN O I-Task
matching NN O I-Task
as NN O O
a NN O O
shape NN O I-Task
matching NN O I-Task
problem NN O I-Task
by NN O O
transforming NN O O
each NN O O
of NN O O
the NN O O
given NN O O
point NN O I-OtherScientificTerm
clouds NN O I-OtherScientificTerm
into NN O O
a NN O O
shape NN O I-Method
representation NN O I-Method
called NN O O
the NN O O
Schrödinger NN O I-Method
distance NN O I-Method
transform NN O I-Method
-LRB- NN O I-Method
SDT NN O I-Method
-RRB- NN O I-Method
representation NN O I-Method
. NN O O

This NN O O
is NN O O
achieved NN O O
by NN O O
solving NN O O
a NN O O
static NN O I-OtherScientificTerm
Schrödinger NN O I-OtherScientificTerm
equation NN O I-OtherScientificTerm
instead NN O O
of NN O O
the NN O O
corresponding NN O O
static NN O I-OtherScientificTerm
Hamilton-Jacobi NN O I-OtherScientificTerm
equation NN O I-OtherScientificTerm
in NN O O
this NN O O
setting NN O O
. NN O O

The NN O O
SDT NN O I-Method
representation NN O I-Method
is NN O O
an NN O O
analytic NN O I-OtherScientificTerm
expression NN O I-OtherScientificTerm
and NN O O
following NN O O
the NN O O
theoretical NN O I-OtherScientificTerm
physics NN O I-OtherScientificTerm
literature NN O I-OtherScientificTerm
, NN O O
can NN O O
be NN O O
normalized NN O O
to NN O O
have NN O O
unit NN O O
L2 NN O O
norm-making NN O O
it NN O I-Generic
a NN O O
square-root NN O I-OtherScientificTerm
density NN O I-OtherScientificTerm
, NN O O
which NN O O
is NN O O
identified NN O O
with NN O O
a NN O O
point NN O O
on NN O O
a NN O O
unit NN O I-OtherScientificTerm
Hilbert NN O I-OtherScientificTerm
sphere NN O I-OtherScientificTerm
, NN O O
whose NN O O
intrinsic NN O I-OtherScientificTerm
geometry NN O I-OtherScientificTerm
is NN O O
fully NN O O
known NN O O
. NN O O

The NN O O
Fisher-Rao NN O I-Metric
metric NN O I-Metric
, NN O O
a NN O O
natural NN O I-Generic
metric NN O I-Generic
for NN O O
the NN O O
space NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
densities NN O I-OtherScientificTerm
leads NN O O
to NN O O
analytic NN O I-OtherScientificTerm
expressions NN O I-OtherScientificTerm
for NN O O
the NN O O
geodesic NN O I-OtherScientificTerm
distance NN O I-OtherScientificTerm
between NN O O
points NN O O
on NN O O
this NN O O
sphere NN O I-Generic
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
use NN O O
the NN O O
well NN O O
known NN O O
Riemannian NN O I-Method
framework NN O I-Method
never NN O O
before NN O O
used NN O O
for NN O O
point NN O I-Task
cloud NN O I-Task
matching NN O I-Task
, NN O O
and NN O O
present NN O O
a NN O O
novel NN O O
matching NN O I-Method
algorithm NN O I-Method
. NN O O

We NN O O
pose NN O O
point NN O I-Task
set NN O I-Task
matching NN O I-Task
under NN O O
rigid NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
non-rigid NN O I-OtherScientificTerm
transformations NN O I-OtherScientificTerm
in NN O O
this NN O O
framework NN O I-Generic
and NN O O
solve NN O O
for NN O O
the NN O O
transformations NN O I-Generic
using NN O O
standard NN O O
nonlinear NN O I-Method
optimization NN O I-Method
techniques NN O I-Method
. NN O O

Finally NN O O
, NN O O
to NN O O
evaluate NN O O
the NN O O
performance NN O O
of NN O O
our NN O O
algorithm-dubbed NN O O
SDTM-we NN O O
present NN O O
several NN O O
synthetic NN O I-Material
and NN O I-Material
real NN O I-Material
data NN O I-Material
examples NN O I-Material
along NN O O
with NN O O
extensive NN O O
comparisons NN O O
to NN O O
state-of-the-art NN O I-Generic
techniques NN O I-Generic
. NN O O

The NN O O
experiments NN O O
show NN O O
that NN O O
our NN O O
algorithm NN O I-Generic
outperforms NN O O
state-of-the-art NN O O
point NN O I-Method
set NN O I-Method
registration NN O I-Method
algorithms NN O I-Method
on NN O O
many NN O O
quantitative NN O I-Metric
metrics NN O I-Metric
. NN O O

-DOCSTART- NN O O

Using NN O O
natural NN O I-Method
language NN O I-Method
processing NN O I-Method
, NN O O
we NN O O
carried NN O O
out NN O O
a NN O O
trend NN O I-Task
survey NN O I-Task
on NN O I-Task
Japanese NN O I-Material
natural NN O I-Material
language NN O I-Material
processing NN O I-Material
studies NN O I-Material
that NN O O
have NN O O
been NN O O
done NN O O
over NN O O
the NN O O
last NN O O
ten NN O O
years NN O O
. NN O O

We NN O O
determined NN O O
the NN O O
changes NN O O
in NN O O
the NN O O
number NN O O
of NN O O
papers NN O O
published NN O O
for NN O O
each NN O O
research NN O O
organization NN O O
and NN O O
on NN O O
each NN O O
research NN O O
area NN O O
as NN O O
well NN O O
as NN O O
the NN O O
relationship NN O O
between NN O O
research NN O O
organizations NN O O
and NN O O
research NN O O
areas NN O O
. NN O O

This NN O O
paper NN O O
is NN O O
useful NN O O
for NN O O
both NN O O
recognizing NN O O
trends NN O O
in NN O O
Japanese NN O I-Material
NLP NN O I-Material
and NN O O
constructing NN O O
a NN O O
method NN O O
of NN O O
supporting NN O O
trend NN O I-Task
surveys NN O I-Task
using NN O O
NLP NN O I-Method
. NN O O

-DOCSTART- NN O O

High NN O I-OtherScientificTerm
frequency NN O I-OtherScientificTerm
oscillations NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
HFOs NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
are NN O O
a NN O O
promising NN O O
biomarker NN O O
of NN O O
epileptic NN O I-OtherScientificTerm
brain NN O I-OtherScientificTerm
tissue NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
activity NN O I-OtherScientificTerm
. NN O O

HFOs NN O I-OtherScientificTerm
additionally NN O O
serve NN O O
as NN O O
a NN O O
prototypical NN O O
example NN O O
of NN O O
challenges NN O O
in NN O O
the NN O O
analysis NN O I-Task
of NN O I-Task
discrete NN O I-Task
events NN O I-Task
in NN O O
high-temporal NN O I-Material
resolution NN O I-Material
, NN O I-Material
intracranial NN O I-Material
EEG NN O I-Material
data NN O I-Material
. NN O O

Two NN O O
primary NN O O
challenges NN O O
are NN O O
1 NN O O
-RRB- NN O O
dimensionality NN O I-Task
reduction NN O I-Task
, NN O O
and NN O O
2 NN O O
-RRB- NN O O
assessing NN O I-Task
feasibility NN O I-Task
of NN O I-Task
classification NN O B-Task
. NN O O

Dimensionality NN O I-Task
reduction NN O I-Task
assumes NN O O
that NN O O
the NN O O
data NN O O
lie NN O O
on NN O O
a NN O O
manifold NN O I-Material
with NN O O
dimension NN O O
less NN O O
than NN O O
that NN O O
of NN O O
the NN O O
features NN O I-OtherScientificTerm
space NN O I-OtherScientificTerm
. NN O O

However NN O O
, NN O O
previous NN O O
HFO NN O I-Task
analysis NN O I-Task
have NN O O
assumed NN O O
a NN O O
linear NN O I-OtherScientificTerm
manifold NN O I-OtherScientificTerm
, NN O O
global NN O O
across NN O O
time NN O O
, NN O O
space NN O O
-LRB- NN O O
i.e. NN O O
recording NN O O
electrode/channel NN O O
-RRB- NN O O
, NN O O
and NN O O
individual NN O O
patients NN O O
. NN O O

Instead NN O O
, NN O O
we NN O O
assess NN O O
both NN O O
a NN O O
-RRB- NN O O
whether NN O O
linear NN O I-Method
methods NN O I-Method
are NN O O
appropriate NN O O
and NN O O
b NN O O
-RRB- NN O O
the NN O O
consistency NN O O
of NN O O
the NN O O
manifold NN O I-OtherScientificTerm
across NN O O
time NN O O
, NN O O
space NN O O
, NN O O
and NN O O
patients NN O O
. NN O O

We NN O O
also NN O O
estimate NN O O
bounds NN O I-OtherScientificTerm
on NN O O
the NN O O
Bayes NN O I-OtherScientificTerm
classification NN O I-OtherScientificTerm
error NN O I-OtherScientificTerm
to NN O O
quantify NN O O
the NN O O
distinction NN O O
between NN O O
two NN O O
classes NN O O
of NN O O
HFOs NN O I-OtherScientificTerm
-LRB- NN O O
those NN O I-Generic
occurring NN O O
during NN O O
seizures NN O O
and NN O O
those NN O I-Generic
occurring NN O O
due NN O O
to NN O O
other NN O O
processes NN O O
-RRB- NN O O
. NN O O

This NN O O
analysis NN O O
provides NN O O
the NN O O
foundation NN O O
for NN O O
future NN O O
clinical NN O I-Task
use NN O I-Task
of NN O I-Task
HFO NN O I-OtherScientificTerm
features NN O I-OtherScientificTerm
and NN O O
guides NN O O
the NN O O
analysis NN O O
for NN O O
other NN O O
discrete NN O I-OtherScientificTerm
events NN O I-OtherScientificTerm
, NN O O
such NN O O
as NN O O
individual NN O O
action NN O I-OtherScientificTerm
potentials NN O I-OtherScientificTerm
or NN O O
multi-unit NN O I-OtherScientificTerm
activity NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
we NN O O
present NN O O
ONTOSCORE NN O I-Method
, NN O O
a NN O O
system NN O I-Generic
for NN O O
scoring NN O O
sets NN O O
of NN O O
concepts NN O O
on NN O O
the NN O O
basis NN O O
of NN O O
an NN O O
ontology NN O I-OtherScientificTerm
. NN O O

We NN O O
apply NN O O
our NN O O
system NN O I-Generic
to NN O O
the NN O O
task NN O O
of NN O O
scoring NN O O
alternative NN O O
speech NN O I-OtherScientificTerm
recognition NN O I-OtherScientificTerm
hypotheses NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
SRH NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
in NN O O
terms NN O O
of NN O O
their NN O O
semantic NN O I-OtherScientificTerm
coherence NN O I-OtherScientificTerm
. NN O O

We NN O O
conducted NN O O
an NN O O
annotation NN O O
experiment NN O O
and NN O O
showed NN O O
that NN O O
human NN O O
annotators NN O O
can NN O O
reliably NN O O
differentiate NN O O
between NN O O
semantically NN O O
coherent NN O O
and NN O O
incoherent NN O O
speech NN O I-OtherScientificTerm
recognition NN O I-OtherScientificTerm
hypotheses NN O I-OtherScientificTerm
. NN O O

An NN O O
evaluation NN O O
of NN O O
our NN O O
system NN O I-Generic
against NN O O
the NN O O
annotated NN O O
data NN O O
shows NN O O
that NN O O
, NN O O
it NN O I-Generic
successfully NN O O
classifies NN O O
73.2 NN O O
% NN O O
in NN O O
a NN O O
German NN O I-Material
corpus NN O I-Material
of NN O O
2.284 NN O O
SRHs NN O I-OtherScientificTerm
as NN O O
either NN O O
coherent NN O O
or NN O O
incoherent NN O O
-LRB- NN O O
given NN O O
a NN O O
baseline NN O O
of NN O O
54.55 NN O O
% NN O O
-RRB- NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
an NN O O
efficient NN O O
dialogue NN O I-Method
management NN O I-Method
for NN O O
an NN O O
information NN O I-Task
navigation NN O I-Task
system NN O I-Task
based NN O O
on NN O O
a NN O O
document NN O I-Material
knowledge NN O I-Material
base NN O I-Material
. NN O O

It NN O O
is NN O O
expected NN O O
that NN O O
incorporation NN O O
of NN O O
appropriate NN O O
N-best NN O I-OtherScientificTerm
candidates NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
ASR NN O I-OtherScientificTerm
and NN O O
contextual NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
will NN O O
improve NN O O
the NN O O
system NN O I-Generic
performance NN O O
. NN O O

The NN O O
system NN O I-Generic
also NN O O
has NN O O
several NN O O
choices NN O O
in NN O O
generating NN O I-Task
responses NN O I-Task
or NN O I-Task
confirmations NN O I-Task
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
this NN O O
selection NN O O
is NN O O
optimized NN O O
as NN O O
minimization NN O I-Task
of NN O I-Task
Bayes NN O I-Task
risk NN O I-Task
based NN O O
on NN O O
reward NN O I-Metric
for NN O O
correct NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
presentation NN O I-OtherScientificTerm
and NN O O
penalty NN O I-Metric
for NN O O
redundant NN O I-OtherScientificTerm
turns NN O I-OtherScientificTerm
. NN O O

We NN O O
have NN O O
evaluated NN O O
this NN O O
strategy NN O I-Generic
with NN O O
our NN O O
spoken NN O I-Task
dialogue NN O I-Task
system NN O I-Task
'' NN O I-Task
Dialogue NN O I-Task
Navigator NN O I-Task
for NN O I-Task
Kyoto NN O I-Task
City NN O I-Task
'' NN O I-Task
, NN O O
which NN O O
also NN O O
has NN O O
question-answering NN O I-OtherScientificTerm
capability NN O I-OtherScientificTerm
. NN O O

Effectiveness NN O O
of NN O O
the NN O O
proposed NN O O
framework NN O I-Generic
was NN O O
confirmed NN O O
in NN O O
the NN O O
success NN O I-Metric
rate NN O I-Metric
of NN O I-Metric
retrieval NN O I-Metric
and NN O O
the NN O O
average NN O I-Metric
number NN O I-Metric
of NN O I-Metric
turns NN O I-Metric
for NN O O
information NN O I-OtherScientificTerm
access NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

Reducing NN O I-Task
language NN O I-Task
model NN O I-Task
-LRB- NN O I-Task
LM NN O I-Task
-RRB- NN O I-Task
size NN O I-Task
is NN O O
a NN O O
critical NN O O
issue NN O O
when NN O O
applying NN O O
a NN O O
LM NN O I-Method
to NN O O
realistic NN O O
applications NN O O
which NN O O
have NN O O
memory NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
. NN O I-OtherScientificTerm

In NN O O
this NN O O
paper NN O O
, NN O O
three NN O O
measures NN O O
are NN O O
studied NN O O
for NN O O
the NN O O
purpose NN O O
of NN O O
LM NN O I-Task
pruning NN O I-Task
. NN O O

They NN O O
are NN O O
probability NN O O
, NN O O
rank NN O I-OtherScientificTerm
, NN O O
and NN O O
entropy NN O I-OtherScientificTerm
. NN O O

We NN O O
evaluated NN O O
the NN O O
performance NN O O
of NN O O
the NN O O
three NN O O
pruning NN O I-OtherScientificTerm
criteria NN O I-OtherScientificTerm
in NN O O
a NN O O
real NN O O
application NN O O
of NN O O
Chinese NN O I-Material
text NN O I-Material
input NN O I-Material
in NN O O
terms NN O O
of NN O O
character NN O I-Metric
error NN O I-Metric
rate NN O I-Metric
-LRB- NN O I-Metric
CER NN O I-Metric
-RRB- NN O I-Metric
. NN O O

We NN O O
first NN O O
present NN O O
an NN O O
empirical NN O O
comparison NN O O
, NN O O
showing NN O O
that NN O O
rank NN O I-OtherScientificTerm
performs NN O O
the NN O O
best NN O O
in NN O O
most NN O O
cases NN O O
. NN O O

We NN O O
also NN O O
show NN O O
that NN O O
the NN O O
high-performance NN O O
of NN O O
rank NN O I-OtherScientificTerm
lies NN O O
in NN O O
its NN O O
strong NN O O
correlation NN O O
with NN O O
error NN O I-Metric
rate NN O I-Metric
. NN O O

We NN O O
then NN O O
present NN O O
a NN O O
novel NN O O
method NN O I-Generic
of NN O O
combining NN O O
two NN O O
criteria NN O O
in NN O O
model NN O I-Task
pruning NN O I-Task
. NN O O

Experimental NN O O
results NN O O
show NN O O
that NN O O
the NN O O
combined NN O O
criterion NN O O
consistently NN O O
leads NN O O
to NN O O
smaller NN O O
models NN O O
than NN O O
the NN O O
models NN O O
pruned NN O O
using NN O O
either NN O O
of NN O O
the NN O O
criteria NN O O
separately NN O O
, NN O O
at NN O O
the NN O O
same NN O O
CER NN O I-Metric
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
proposes NN O O
an NN O O
annotating NN O I-Method
scheme NN O I-Method
that NN O O
encodes NN O O
honorifics NN O I-Material
-LRB- NN O O
respectful NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
-RRB- NN O O
. NN O O

Honorifics NN O I-Material
are NN O O
used NN O O
extensively NN O O
in NN O O
Japanese NN O I-Material
, NN O O
reflecting NN O O
the NN O O
social NN O O
relationship NN O O
-LRB- NN O O
e.g. NN O O
social NN O O
ranks NN O O
and NN O O
age NN O O
-RRB- NN O O
of NN O O
the NN O O
referents NN O O
. NN O O

This NN O O
referential NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
is NN O O
vital NN O O
for NN O O
resolving NN O O
zero NN O I-OtherScientificTerm
pronouns NN O I-OtherScientificTerm
and NN O O
improving NN O O
machine NN O I-OtherScientificTerm
translation NN O I-OtherScientificTerm
outputs NN O I-OtherScientificTerm
. NN O O

Annotating NN O O
honorifics NN O I-Material
is NN O O
a NN O O
complex NN O O
task NN O O
that NN O O
involves NN O O
identifying NN O O
a NN O O
predicate NN O O
with NN O O
honorifics NN O I-Material
, NN O O
assigning NN O O
ranks NN O I-OtherScientificTerm
to NN O O
referents NN O O
of NN O O
the NN O O
predicate NN O O
, NN O O
calibrating NN O O
the NN O O
ranks NN O I-OtherScientificTerm
, NN O O
and NN O O
connecting NN O O
referents NN O O
with NN O O
their NN O O
predicates NN O O
. NN O O

-DOCSTART- NN O O

Visually-guided NN O I-OtherScientificTerm
arm NN O I-OtherScientificTerm
reaching NN O I-OtherScientificTerm
movements NN O I-OtherScientificTerm
are NN O O
produced NN O O
by NN O O
distributed NN O I-Method
neural NN O I-Method
networks NN O I-Method
within NN O O
parietal NN O O
and NN O O
frontal NN O O
regions NN O O
of NN O O
the NN O O
cerebral NN O O
cortex NN O O
. NN O O

Experimental NN O O
data NN O O
indicate NN O O
that NN O O
-LRB- NN O O
I NN O O
-RRB- NN O O
single NN O O
neurons NN O O
in NN O O
these NN O O
regions NN O O
are NN O O
broadly NN O O
tuned NN O O
to NN O O
parameters NN O O
of NN O O
movement NN O O
; NN O O
-LRB- NN O O
2 NN O O
-RRB- NN O O
appropriate NN O O
commands NN O O
are NN O O
elaborated NN O O
by NN O O
populations NN O O
of NN O O
neurons NN O O
; NN O O
-LRB- NN O O
3 NN O O
-RRB- NN O O
the NN O O
coordinated NN O I-OtherScientificTerm
action NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
neu-rons NN O I-OtherScientificTerm
can NN O O
be NN O O
visualized NN O O
using NN O O
a NN O O
neuronal NN O I-OtherScientificTerm
population NN O I-OtherScientificTerm
vector NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
NPV NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
. NN O O

However NN O O
, NN O O
the NN O O
NPV NN O I-OtherScientificTerm
provides NN O O
only NN O O
a NN O O
rough NN O O
estimate NN O O
of NN O O
movement NN O O
parameters NN O O
-LRB- NN O O
direction NN O O
, NN O O
velocity NN O O
-RRB- NN O O
and NN O O
may NN O O
even NN O O
fail NN O O
to NN O O
reflect NN O O
the NN O O
parameters NN O O
of NN O O
movement NN O O
when NN O O
arm NN O I-OtherScientificTerm
posture NN O I-OtherScientificTerm
is NN O O
changed NN O O
. NN O O

We NN O O
designed NN O O
a NN O O
model NN O I-Generic
of NN O O
the NN O O
cortical NN O I-OtherScientificTerm
motor NN O I-OtherScientificTerm
command NN O I-OtherScientificTerm
to NN O O
investigate NN O O
the NN O O
relation NN O O
between NN O O
the NN O O
desired NN O O
direction NN O O
of NN O O
the NN O O
movement NN O O
, NN O O
the NN O O
actual NN O O
direction NN O O
of NN O O
movement NN O O
and NN O O
the NN O O
direction NN O O
of NN O O
the NN O O
NPV NN O I-OtherScientificTerm
in NN O O
motor NN O I-OtherScientificTerm
cortex NN O I-OtherScientificTerm
. NN O O

The NN O O
model NN O I-Generic
is NN O O
a NN O O
two-layer NN O I-Method
self-organizing NN O I-Method
neural NN O I-Method
network NN O I-Method
which NN O O
combines NN O O
broadly-tuned NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
muscular NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
proprioceptive NN O I-OtherScientificTerm
and NN O O
-LRB- NN O I-OtherScientificTerm
cartesian NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
visual NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
to NN O O
calculate NN O O
-LRB- NN O I-OtherScientificTerm
angular NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
motor NN O I-OtherScientificTerm
commands NN O I-OtherScientificTerm
for NN O O
the NN O O
initial NN O O
part NN O O
of NN O O
the NN O O
movement NN O O
of NN O O
a NN O O
two-link NN O I-OtherScientificTerm
arm NN O I-OtherScientificTerm
. NN O O

The NN O O
network NN O I-Generic
was NN O O
trained NN O O
by NN O O
motor NN O O
babbling NN O O
in NN O O
5 NN O O
positions NN O O
. NN O O

Simulations NN O O
showed NN O O
that NN O O
-LRB- NN O O
1 NN O O
-RRB- NN O O
the NN O O
network NN O I-Generic
produced NN O O
appropriate NN O O
movement NN O O
direction NN O O
over NN O O
a NN O O
large NN O O
part NN O O
of NN O O
the NN O O
workspace NN O O
; NN O O
-LRB- NN O O
2 NN O O
-RRB- NN O O
small NN O O
deviations NN O O
of NN O O
the NN O O
actual NN O O
trajectory NN O O
from NN O O
the NN O O
desired NN O O
trajectory NN O O
existed NN O O
at NN O O
the NN O O
extremities NN O O
of NN O O
the NN O O
workspace NN O O
; NN O O
-LRB- NN O O
3 NN O O
-RRB- NN O O
these NN O O
deviations NN O O
were NN O O
accompanied NN O O
by NN O O
large NN O O
deviations NN O O
of NN O O
the NN O O
NPV NN O I-OtherScientificTerm
from NN O O
both NN O O
trajectories NN O O
. NN O O

These NN O O
results NN O O
suggest NN O O
the NN O O
NPV NN O I-OtherScientificTerm
does NN O O
not NN O O
give NN O O
a NN O O
faithful NN O O
image NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
cortical NN O I-OtherScientificTerm
processing NN O I-OtherScientificTerm
during NN O O
arm NN O I-OtherScientificTerm
reaching NN O I-OtherScientificTerm
movements NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

It NN O O
is NN O O
well-known NN O O
that NN O O
diversity NN O O
among NN O O
base NN O I-Method
classifiers NN O I-Method
is NN O O
crucial NN O O
for NN O O
constructing NN O O
a NN O O
strong NN O O
ensemble NN O I-Task
. NN O O

Most NN O O
existing NN O O
ensemble NN O I-Method
methods NN O I-Method
obtain NN O O
diverse NN O O
individual NN O O
learners NN O O
through NN O O
resampling NN O O
the NN O O
instances NN O O
or NN O O
features NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
an NN O O
alternative NN O O
way NN O O
for NN O O
ensemble NN O I-Task
construction NN O I-Task
by NN O O
resampling NN O I-Task
pairwise NN O I-Task
constraints NN O I-Task
that NN O O
specify NN O O
whether NN O O
a NN O O
pair NN O O
of NN O O
instances NN O O
belongs NN O O
to NN O O
the NN O O
same NN O O
class NN O O
or NN O O
not NN O O
. NN O O

Using NN O O
pairwise NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
for NN O O
ensemble NN O I-Task
construction NN O I-Task
is NN O O
challenging NN O O
because NN O O
it NN O O
remains NN O O
unknown NN O O
how NN O O
to NN O O
influence NN O O
the NN O O
base NN O I-Method
classifiers NN O I-Method
with NN O O
the NN O O
sampled NN O I-OtherScientificTerm
pairwise NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
. NN O O

We NN O O
solve NN O O
this NN O O
problem NN O O
with NN O O
a NN O O
two-step NN O O
process NN O O
. NN O O

First NN O O
, NN O O
we NN O O
transform NN O O
the NN O O
original NN O O
instances NN O O
into NN O O
a NN O O
new NN O O
data NN O I-Method
representation NN O I-Method
using NN O O
projections NN O I-OtherScientificTerm
learnt NN O O
from NN O O
pairwise NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
. NN O O

Then NN O O
, NN O O
we NN O O
build NN O O
the NN O O
base NN O I-Method
clas-sifiers NN O I-Method
with NN O O
the NN O O
new NN O O
data NN O I-Method
representation NN O I-Method
. NN O O

We NN O O
propose NN O O
two NN O O
methods NN O O
for NN O O
resampling NN O I-Task
pairwise NN O I-Task
constraints NN O I-Task
following NN O O
the NN O O
standard NN O O
Bagging NN O I-Method
and NN O I-Method
Boosting NN O I-Method
algorithms NN O I-Method
, NN O O
respectively NN O O
. NN O O

Extensive NN O O
experiments NN O O
validate NN O O
the NN O O
effectiveness NN O O
of NN O O
our NN O O
method NN O O
. NN O O

-DOCSTART- NN O O

A NN O O
new NN O O
algorithm NN O I-Generic
for NN O O
solving NN O O
the NN O O
three NN O O
dimensional NN O I-Task
container NN O I-Task
packing NN O I-Task
problem NN O I-Task
is NN O O
proposed NN O O
in NN O O
this NN O O
paper NN O O
. NN O O

This NN O O
new NN O O
algorithm NN O I-Generic
deviates NN O O
from NN O O
the NN O O
traditional NN O O
approach NN O I-Method
of NN O I-Method
wall NN O I-Method
building NN O I-Method
and NN O I-Method
layering NN O I-Method
. NN O O

It NN O I-Generic
uses NN O O
the NN O O
concept NN O O
of NN O O
'' NN O O
building NN O O
growing NN O O
'' NN O O
from NN O O
multiple NN O O
sides NN O O
of NN O O
the NN O O
container NN O O
. NN O O

We NN O O
tested NN O O
our NN O O
method NN O I-Generic
using NN O O
all NN O O
760 NN O O
test NN O O
cases NN O O
from NN O O
the NN O O
OR-Library NN O I-Material
. NN O O

Experimental NN O O
results NN O O
indicate NN O O
that NN O O
the NN O O
new NN O O
algorithm NN O I-Generic
is NN O O
able NN O O
to NN O O
achieve NN O O
an NN O O
average NN O I-Metric
packing NN O I-Metric
utilization NN O I-Metric
of NN O O
more NN O O
than NN O O
87 NN O O
% NN O O
. NN O O

This NN O O
is NN O O
better NN O O
than NN O O
the NN O O
results NN O O
reported NN O O
in NN O O
the NN O O
literature NN O O
. NN O O

-DOCSTART- NN O O

Current NN O O
approaches NN O I-Generic
to NN O O
object NN O I-Task
category NN O I-Task
recognition NN O I-Task
require NN O O
datasets NN O I-Generic
of NN O O
training NN O O
images NN O O
to NN O O
be NN O O
manually NN O O
prepared NN O O
, NN O O
with NN O O
varying NN O O
degrees NN O O
of NN O O
supervision NN O I-OtherScientificTerm
. NN O O

We NN O O
present NN O O
an NN O O
approach NN O I-Generic
that NN O O
can NN O O
learn NN O O
an NN O O
object NN O I-OtherScientificTerm
category NN O I-OtherScientificTerm
from NN O O
just NN O O
its NN O O
name NN O O
, NN O O
by NN O O
utilizing NN O O
the NN O O
raw NN O O
output NN O O
of NN O O
image NN O I-Method
search NN O I-Method
engines NN O I-Method
available NN O O
on NN O O
the NN O O
Internet NN O O
. NN O O

We NN O O
develop NN O O
a NN O O
new NN O O
model NN O I-Generic
, NN O O
TSI-pLSA NN O I-Method
, NN O O
which NN O O
extends NN O O
pLSA NN O I-Method
-LRB- NN O O
as NN O O
applied NN O O
to NN O O
visual NN O I-OtherScientificTerm
words NN O I-OtherScientificTerm
-RRB- NN O O
to NN O O
include NN O O
spatial NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
in NN O O
a NN O O
translation NN O O
and NN O O
scale NN O O
invariant NN O O
manner NN O O
. NN O O

Our NN O O
approach NN O I-Generic
can NN O O
handle NN O O
the NN O O
high NN O O
intra-class NN O I-OtherScientificTerm
variability NN O I-OtherScientificTerm
and NN O O
large NN O O
proportion NN O O
of NN O O
unrelated NN O I-OtherScientificTerm
images NN O I-OtherScientificTerm
returned NN O O
by NN O O
search NN O I-Method
engines NN O I-Method
. NN O O

We NN O O
evaluate NN O O
the NN O O
models NN O I-Generic
on NN O O
standard NN O O
test NN O I-Generic
sets NN O I-Generic
, NN O O
showing NN O O
performance NN O O
competitive NN O O
with NN O O
existing NN O O
methods NN O I-Generic
trained NN O O
on NN O O
hand NN O I-Material
prepared NN O I-Material
datasets NN O I-Material
. NN O O

-DOCSTART- NN O O

The NN O O
paper NN O O
provides NN O O
an NN O O
overview NN O O
of NN O O
the NN O O
research NN O O
conducted NN O O
at NN O O
LIMSI NN O O
in NN O O
the NN O O
field NN O O
of NN O O
speech NN O I-Task
processing NN O I-Task
, NN O O
but NN O O
also NN O O
in NN O O
the NN O O
related NN O O
areas NN O O
of NN O O
Human-Machine NN O I-Task
Communication NN O I-Task
, NN O O
including NN O O
Natural NN O I-Task
Language NN O I-Task
Processing NN O I-Task
, NN O O
Non NN O I-Task
Verbal NN O I-Task
and NN O I-Task
Multimodal NN O I-Task
Communication NN O I-Task
. NN O O

Also NN O O
presented NN O O
are NN O O
the NN O O
commercial NN O O
applications NN O O
of NN O O
some NN O O
of NN O O
the NN O O
research NN O O
projects NN O O
. NN O O

When NN O O
applicable NN O O
, NN O O
the NN O O
discussion NN O O
is NN O O
placed NN O O
in NN O O
the NN O O
framework NN O O
of NN O O
international NN O O
collaborations NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
have NN O O
calculated NN O O
analytical NN O I-OtherScientificTerm
expressions NN O I-OtherScientificTerm
for NN O O
how NN O O
the NN O O
bias NN O O
and NN O O
variance NN O O
of NN O O
the NN O O
estimators NN O O
provided NN O O
by NN O O
various NN O O
temporal NN O I-Method
difference NN O I-Method
value NN O I-Method
estimation NN O I-Method
algorithms NN O I-Method
change NN O O
with NN O O
offline NN O O
updates NN O O
over NN O O
trials NN O O
in NN O O
absorbing NN O O
Markov NN O I-OtherScientificTerm
chains NN O I-OtherScientificTerm
using NN O O
lookup NN O I-Method
table NN O I-Method
representations NN O I-Method
. NN O O

We NN O O
illustrate NN O O
classes NN O O
of NN O O
learning NN O I-OtherScientificTerm
curve NN O I-OtherScientificTerm
behavior NN O I-OtherScientificTerm
in NN O O
various NN O O
chains NN O O
, NN O O
and NN O O
show NN O O
the NN O O
manner NN O O
in NN O O
which NN O O
TD NN O O
is NN O O
sensitive NN O O
to NN O O
the NN O O
choice NN O O
of NN O O
its NN O O
step-size NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
eligibility NN O I-OtherScientificTerm
trace NN O I-OtherScientificTerm
parameters NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
describe NN O O
the NN O O
pronominal NN O I-Method
anaphora NN O I-Method
resolution NN O I-Method
module NN O I-Method
of NN O O
Lucy NN O I-Method
, NN O O
a NN O O
portable NN O O
English NN O I-Method
understanding NN O I-Method
system NN O I-Method
. NN O O

The NN O O
design NN O O
of NN O O
this NN O O
module NN O I-Generic
was NN O O
motivated NN O O
by NN O O
the NN O O
observation NN O O
that NN O O
, NN O O
although NN O O
there NN O O
exist NN O O
many NN O O
theories NN O O
of NN O O
anaphora NN O I-Task
resolution NN O I-Task
, NN O O
no NN O O
one NN O O
of NN O O
these NN O O
theories NN O O
is NN O O
complete NN O O
. NN O O

Thus NN O O
we NN O O
have NN O O
implemented NN O O
a NN O O
blackboard-like NN O I-Method
architecture NN O I-Method
in NN O O
which NN O O
individual NN O O
partial NN O O
theories NN O O
can NN O O
be NN O O
encoded NN O O
as NN O O
separate NN O O
modules NN O O
that NN O O
can NN O O
interact NN O O
to NN O O
propose NN O O
candidate NN O O
antecedents NN O O
and NN O O
to NN O O
evaluate NN O O
each NN O O
other NN O O
's NN O O
proposals NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
reported NN O O
experiments NN O O
of NN O O
unsupervised NN O I-Task
automatic NN O I-Task
acquisition NN O I-Task
of NN O I-Task
Italian NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
English NN O I-OtherScientificTerm
verb NN O I-OtherScientificTerm
subcategorization NN O I-OtherScientificTerm
frames NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
SCFs NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
from NN O O
general NN O I-Material
and NN O I-Material
domain NN O I-Material
corpora NN O I-Material
. NN O O

The NN O O
proposed NN O O
technique NN O I-Generic
operates NN O O
on NN O O
syntactically NN O I-Material
shallow-parsed NN O I-Material
corpora NN O I-Material
on NN O O
the NN O O
basis NN O O
of NN O O
a NN O O
limited NN O O
number NN O O
of NN O O
search NN O I-Method
heuristics NN O I-Method
not NN O O
relying NN O O
on NN O O
any NN O O
previous NN O O
lexico-syntactic NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
about NN O O
SCFs NN O I-OtherScientificTerm
. NN O O

Although NN O O
preliminary NN O O
, NN O O
reported NN O O
results NN O O
are NN O O
in NN O O
line NN O O
with NN O O
state-of-the-art NN O O
lexical NN O I-Method
acquisition NN O I-Method
systems NN O I-Method
. NN O O

The NN O O
issue NN O O
of NN O O
whether NN O O
verbs NN O O
sharing NN O O
similar NN O O
SCFs NN O I-OtherScientificTerm
distributions NN O I-OtherScientificTerm
happen NN O O
to NN O O
share NN O O
similar NN O I-OtherScientificTerm
semantic NN O I-OtherScientificTerm
properties NN O I-OtherScientificTerm
as NN O O
well NN O O
was NN O O
also NN O O
explored NN O O
by NN O O
clustering NN O O
verbs NN O O
that NN O O
share NN O O
frames NN O O
with NN O O
the NN O O
same NN O O
distribution NN O O
using NN O O
the NN O O
Minimum NN O I-Method
Description NN O I-Method
Length NN O I-Method
Principle NN O I-Method
-LRB- NN O I-Method
MDL NN O I-Method
-RRB- NN O I-Method
. NN O O

First NN O O
experiments NN O O
in NN O O
this NN O O
direction NN O O
were NN O O
carried NN O O
out NN O O
on NN O O
Italian NN O O
verbs NN O O
with NN O O
encouraging NN O O
results NN O O
. NN O O

-DOCSTART- NN O O

Graph-cuts NN O I-Method
optimization NN O I-Method
is NN O O
prevalent NN O O
in NN O O
vision NN O I-Task
and NN O I-Task
graphics NN O I-Task
problems NN O I-Task
. NN O O

It NN O O
is NN O O
thus NN O O
of NN O O
great NN O O
practical NN O O
importance NN O O
to NN O O
parallelize NN O O
the NN O O
graph-cuts NN O I-Method
optimization NN O I-Method
using NN O O
to-day NN O O
's NN O O
ubiquitous NN O O
multi-core NN O I-OtherScientificTerm
machines NN O I-OtherScientificTerm
. NN O O

However NN O O
, NN O O
the NN O O
current NN O O
best NN O O
serial NN O I-Method
algorithm NN O I-Method
by NN O O
Boykov NN O O
and NN O O
Kolmogorov NN O O
-LSB- NN O O
4 NN O O
-RSB- NN O O
-LRB- NN O O
called NN O O
the NN O O
BK NN O I-Method
algorithm NN O I-Method
-RRB- NN O O
still NN O O
has NN O O
the NN O O
superior NN O O
empirical NN O O
performance NN O O
. NN O O

It NN O O
is NN O O
non-trivial NN O O
to NN O O
parallelize NN O O
as NN O O
expensive NN O O
synchronization NN O I-OtherScientificTerm
overhead NN O I-OtherScientificTerm
easily NN O O
offsets NN O O
the NN O O
advantage NN O O
of NN O O
parallelism NN O I-OtherScientificTerm
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
novel NN O O
adaptive NN O I-Method
bottom-up NN O I-Method
approach NN O I-Method
to NN O O
parallelize NN O O
the NN O O
BK NN O I-Method
algorithm NN O I-Method
. NN O O

We NN O O
first NN O O
uniformly NN O O
partition NN O O
the NN O O
graph NN O I-OtherScientificTerm
into NN O O
a NN O O
number NN O O
of NN O O
regularly-shaped NN O I-OtherScientificTerm
dis-joint NN O I-OtherScientificTerm
subgraphs NN O I-OtherScientificTerm
and NN O O
process NN O O
them NN O I-Generic
in NN O O
parallel NN O O
, NN O O
then NN O O
we NN O O
incre-mentally NN O O
merge NN O O
the NN O O
subgraphs NN O I-OtherScientificTerm
in NN O O
an NN O O
adaptive NN O O
way NN O O
to NN O O
obtain NN O O
the NN O O
global NN O I-OtherScientificTerm
optimum NN O I-OtherScientificTerm
. NN O O

The NN O O
new NN O O
algorithm NN O I-Generic
has NN O O
three NN O O
benefits NN O O
: NN O O
1 NN O O
-RRB- NN O O
it NN O O
is NN O O
more NN O O
cache-friendly NN O O
within NN O O
smaller NN O O
subgraphs NN O I-OtherScientificTerm
; NN O O
2 NN O O
-RRB- NN O O
it NN O O
keeps NN O O
balanced NN O I-OtherScientificTerm
workloads NN O I-OtherScientificTerm
among NN O O
computing NN O O
cores NN O O
; NN O O
3 NN O O
-RRB- NN O O
it NN O O
causes NN O O
little NN O O
overhead NN O I-OtherScientificTerm
and NN O O
is NN O O
adaptable NN O O
to NN O O
the NN O O
number NN O O
of NN O O
available NN O O
cores NN O O
. NN O O

Extensive NN O O
experiments NN O O
in NN O O
common NN O O
applications NN O I-Generic
such NN O O
as NN O O
2D/3D NN O I-Task
image NN O I-Task
segmentations NN O I-Task
and NN O O
3D NN O I-Task
surface NN O I-Task
fitting NN O I-Task
demonstrate NN O O
the NN O O
effectiveness NN O O
of NN O O
our NN O O
approach NN O I-Generic
. NN O O

-DOCSTART- NN O O

We NN O O
study NN O O
the NN O O
question NN O O
of NN O O
how NN O O
to NN O O
make NN O O
loss-aware NN O I-Task
predictions NN O I-Task
in NN O I-Task
image NN O I-Task
segmentation NN O I-Task
settings NN O I-Task
where NN O O
the NN O O
evaluation NN O I-OtherScientificTerm
function NN O I-OtherScientificTerm
is NN O O
the NN O O
Intersection-over-Union NN O I-Metric
-LRB- NN O I-Metric
IoU NN O I-Metric
-RRB- NN O I-Metric
measure NN O I-Metric
that NN O O
is NN O O
used NN O O
widely NN O O
in NN O O
evaluating NN O O
image NN O I-Method
segmentation NN O I-Method
systems NN O I-Method
. NN O O

Currently NN O O
, NN O O
there NN O O
are NN O O
two NN O O
dominant NN O I-Generic
approaches NN O I-Generic
: NN O O
the NN O O
first NN O I-Generic
approximates NN O O
the NN O O
Expected-IoU NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
EIoU NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
score NN O I-OtherScientificTerm
as NN O O
Expected-Intersection-over-Expected-Union NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
EIoEU NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
; NN O O
and NN O O
the NN O O
second NN O I-Generic
approach NN O I-Generic
is NN O O
to NN O O
compute NN O O
exact NN O O
EIoU NN O I-OtherScientificTerm
but NN O O
only NN O O
over NN O O
a NN O O
small NN O O
set NN O O
of NN O O
high-quality NN O O
candidate NN O O
solutions NN O O
. NN O O

We NN O O
begin NN O O
by NN O O
asking NN O O
which NN O O
approach NN O O
we NN O O
should NN O O
favor NN O O
for NN O O
two NN O O
typical NN O O
image NN O I-Task
seg-mentation NN O I-Task
tasks NN O I-Task
. NN O O

Studying NN O O
this NN O O
question NN O O
leads NN O O
to NN O O
two NN O O
new NN O O
methods NN O O
that NN O O
draw NN O O
ideas NN O O
from NN O O
both NN O O
existing NN O O
approaches NN O O
. NN O O

Our NN O O
new NN O O
methods NN O I-Generic
use NN O O
the NN O O
EIoEU NN O I-Method
approximation NN O I-Method
paired NN O O
with NN O O
high NN O O
quality NN O O
candidate NN O O
solutions NN O O
. NN O O

Experimentally NN O O
we NN O O
show NN O O
that NN O O
our NN O O
new NN O O
approaches NN O I-Generic
lead NN O O
to NN O O
improved NN O O
performance NN O O
on NN O O
both NN O O
image NN O I-Task
segmentation NN O I-Task
tasks NN O I-Task
. NN O O

-DOCSTART- NN O O

Boosting NN O I-Method
methods NN O I-Method
are NN O O
known NN O O
not NN O O
to NN O O
usually NN O O
overfit NN O O
training NN O O
data NN O O
even NN O O
as NN O O
the NN O O
size NN O O
of NN O O
the NN O O
generated NN O O
classifiers NN O I-Method
becomes NN O O
large NN O O
. NN O O

Schapire NN O O
et NN O O
al. NN O O
attempted NN O O
to NN O O
explain NN O O
this NN O O
phenomenon NN O O
in NN O O
terms NN O O
of NN O O
the NN O O
margins NN O O
the NN O O
classifier NN O I-Method
achieves NN O O
on NN O O
training NN O O
examples NN O O
. NN O O

Later NN O O
, NN O O
however NN O O
, NN O O
Breiman NN O O
cast NN O O
serious NN O O
doubt NN O O
on NN O O
this NN O O
explanation NN O O
by NN O O
introducing NN O O
a NN O O
boosting NN O I-Method
algorithm NN O I-Method
, NN O O
arc-gv NN O I-Method
, NN O O
that NN O O
can NN O O
generate NN O O
a NN O O
higher NN O O
margins NN O I-OtherScientificTerm
distribution NN O I-OtherScientificTerm
than NN O O
AdaBoost NN O I-Method
and NN O O
yet NN O O
performs NN O O
worse NN O O
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
take NN O O
a NN O O
close NN O O
look NN O O
at NN O O
Breiman NN O O
's NN O O
compelling NN O O
but NN O O
puzzling NN O O
results NN O O
. NN O O

Although NN O O
we NN O O
can NN O O
reproduce NN O O
his NN O O
main NN O O
finding NN O O
, NN O O
we NN O O
find NN O O
that NN O O
the NN O O
poorer NN O O
performance NN O O
of NN O O
arc-gv NN O I-Method
can NN O O
be NN O O
explained NN O O
by NN O O
the NN O O
increased NN O O
complexity NN O I-Metric
of NN O O
the NN O O
base NN O I-Method
classifiers NN O I-Method
it NN O O
uses NN O O
, NN O O
an NN O O
explanation NN O O
supported NN O O
by NN O O
our NN O O
experiments NN O O
and NN O O
entirely NN O O
consistent NN O O
with NN O O
the NN O O
margins NN O I-Method
theory NN O I-Method
. NN O O

Thus NN O O
, NN O O
we NN O O
find NN O O
maximizing NN O O
the NN O O
margins NN O O
is NN O O
desirable NN O O
, NN O O
but NN O O
not NN O O
necessarily NN O O
at NN O O
the NN O O
expense NN O O
of NN O O
other NN O O
factors NN O O
, NN O O
especially NN O O
base-classifier NN O I-Metric
complexity NN O I-Metric
. NN O O

-DOCSTART- NN O O

The NN O O
transfer NN O I-Task
phase NN O I-Task
in NN O O
machine NN O I-Task
translation NN O I-Task
-LRB- NN O I-Task
MT NN O I-Task
-RRB- NN O I-Task
systems NN O I-Task
has NN O O
been NN O O
considered NN O O
to NN O O
be NN O O
more NN O O
complicated NN O O
than NN O O
analysis NN O I-Task
and NN O O
generation NN O I-Task
, NN O O
since NN O O
it NN O I-Generic
is NN O O
inherently NN O O
a NN O O
conglomeration NN O O
of NN O O
individual NN O O
lexical NN O I-OtherScientificTerm
rules NN O I-OtherScientificTerm
. NN O O

Currently NN O O
some NN O O
attempts NN O O
are NN O O
being NN O O
made NN O O
to NN O O
use NN O O
case-based NN O I-Method
reasoning NN O I-Method
in NN O O
machine NN O I-Task
translation NN O I-Task
, NN O O
that NN O O
is NN O O
, NN O O
to NN O O
make NN O O
decisions NN O O
on NN O O
the NN O O
basis NN O O
of NN O O
translation NN O O
examples NN O O
at NN O O
appropriate NN O O
pints NN O O
in NN O O
MT NN O I-Task
. NN O O

This NN O O
paper NN O O
proposes NN O O
a NN O O
new NN O O
type NN O O
of NN O O
transfer NN O I-Method
system NN O I-Method
, NN O O
called NN O O
a NN O O
Similarity-driven NN O I-Method
Transfer NN O I-Method
System NN O I-Method
-LRB- NN O I-Method
SimTran NN O I-Method
-RRB- NN O I-Method
, NN O O
for NN O O
use NN O O
in NN O O
such NN O O
case-based NN O I-Task
MT NN O I-Task
-LRB- NN O I-Task
CBMT NN O I-Task
-RRB- NN O I-Task
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
addresses NN O O
the NN O O
problem NN O O
of NN O O
optimal NN O I-Task
alignment NN O I-Task
of NN O I-Task
non-rigid NN O I-Task
surfaces NN O I-Task
from NN O O
multi-view NN O I-OtherScientificTerm
video NN O I-OtherScientificTerm
observations NN O I-OtherScientificTerm
to NN O O
obtain NN O O
a NN O O
temporally NN O I-Method
consistent NN O I-Method
representation NN O I-Method
. NN O O

Conventional NN O O
non-rigid NN O I-Method
surface NN O I-Method
tracking NN O I-Method
performs NN O O
frame-to-frame NN O I-Method
alignment NN O I-Method
which NN O O
is NN O O
subject NN O O
to NN O O
the NN O O
accumulation NN O O
of NN O O
errors NN O O
resulting NN O O
in NN O O
a NN O O
drift NN O O
over NN O O
time NN O O
. NN O O

Recently NN O O
, NN O O
non-sequential NN O I-Method
tracking NN O I-Method
approaches NN O I-Method
have NN O O
been NN O O
introduced NN O O
which NN O O
reorder NN O O
the NN O O
input NN O O
data NN O O
based NN O O
on NN O O
a NN O O
dissimilarity NN O I-Metric
measure NN O I-Metric
. NN O O

One NN O O
or NN O O
more NN O O
input NN O O
sequences NN O O
are NN O O
represented NN O O
in NN O O
a NN O O
tree NN O O
with NN O O
reduced NN O O
alignment NN O O
path NN O O
length NN O O
. NN O O

They NN O O
demonstrate NN O O
a NN O O
reduced NN O I-OtherScientificTerm
drift NN O I-OtherScientificTerm
and NN O O
increased NN O O
robustness NN O I-Metric
to NN O O
large NN O O
non-rigid NN O I-OtherScientificTerm
deformations NN O I-OtherScientificTerm
. NN O O

However NN O O
, NN O O
jumps NN O O
may NN O O
occur NN O O
in NN O O
the NN O O
aligned NN O O
mesh NN O O
sequence NN O O
where NN O O
branches NN O I-OtherScientificTerm
of NN O O
the NN O O
tree NN O I-OtherScientificTerm
meet NN O O
due NN O O
to NN O O
different NN O O
error NN O I-OtherScientificTerm
accumulation NN O I-OtherScientificTerm
along NN O O
them NN O O
. NN O O

Optimisation NN O I-Task
of NN O I-Task
the NN O I-Task
tree NN O I-Task
for NN O O
non-sequential NN O I-Method
tracking NN O I-Method
, NN O O
which NN O O
minimises NN O O
the NN O O
errors NN O O
in NN O O
temporal NN O I-Metric
consistency NN O I-Metric
due NN O O
to NN O O
both NN O O
the NN O O
drift NN O O
and NN O O
the NN O O
jumps NN O O
, NN O O
is NN O O
proposed NN O O
. NN O O

A NN O O
novel NN O O
cluster NN O I-Method
tree NN O I-Method
enforces NN O O
sequential NN O I-Task
tracking NN O I-Task
in NN O I-Task
local NN O I-OtherScientificTerm
segments NN O I-OtherScientificTerm
of NN O O
the NN O O
sequence NN O O
while NN O O
allowing NN O O
global NN O I-OtherScientificTerm
non-sequential NN O I-OtherScientificTerm
traversal NN O I-OtherScientificTerm
among NN O O
these NN O O
segments NN O I-Generic
. NN O O

This NN O O
provides NN O O
a NN O O
mechanism NN O O
to NN O O
create NN O O
a NN O O
tree NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
which NN O O
reduces NN O O
the NN O O
number NN O O
and NN O O
size NN O O
of NN O O
jumps NN O O
between NN O O
branches NN O O
and NN O O
limits NN O O
the NN O O
lenght NN O O
of NN O O
branches NN O O
as NN O O
well NN O O
. NN O O

Comprehensive NN O O
evaluation NN O O
is NN O O
performed NN O O
on NN O O
a NN O O
variety NN O O
of NN O O
challenging NN O O
non-rigid NN O I-OtherScientificTerm
surfaces NN O I-OtherScientificTerm
including NN O O
face NN O I-OtherScientificTerm
, NN O O
cloth NN O I-OtherScientificTerm
and NN O O
people NN O I-OtherScientificTerm
. NN O O

It NN O O
demonstrates NN O O
that NN O O
the NN O O
proposed NN O O
cluster NN O I-Method
tree NN O I-Method
achieves NN O O
better NN O O
temporal NN O I-Metric
consistency NN O I-Metric
than NN O O
the NN O O
previous NN O O
sequential NN O I-Method
and NN O I-Method
non-sequential NN O I-Method
tracking NN O I-Method
approaches NN O I-Method
. NN O O

Quantitative NN O O
analysis NN O O
on NN O O
a NN O O
created NN O O
synthetic NN O I-Metric
facial NN O I-Metric
performance NN O I-Metric
also NN O O
shows NN O O
an NN O O
improvement NN O O
by NN O O
the NN O O
cluster NN O I-Method
tree NN O I-Method
. NN O O

-DOCSTART- NN O O

The NN O O
translation NN O I-Task
of NN O I-Task
English NN O I-Task
text NN O I-Task
into NN O I-Task
American NN O I-Task
Sign NN O I-Task
Language NN O I-Task
-LRB- NN O I-Task
ASL NN O I-Task
-RRB- NN O I-Task
animation NN O I-Task
tests NN O O
the NN O O
limits NN O O
of NN O O
traditional NN O O
MT NN O I-Method
architectural NN O I-Method
designs NN O I-Method
. NN O O

A NN O O
new NN O O
semantic NN O I-Method
representation NN O I-Method
is NN O O
proposed NN O O
that NN O O
uses NN O O
virtual NN O I-Method
reality NN O I-Method
3D NN O I-Method
scene NN O I-Method
modeling NN O I-Method
software NN O I-Method
to NN O O
produce NN O O
spatially NN O I-OtherScientificTerm
complex NN O I-OtherScientificTerm
ASL NN O I-OtherScientificTerm
phenomena NN O I-OtherScientificTerm
called NN O O
'' NN O O
classifier NN O I-OtherScientificTerm
predicates NN O I-OtherScientificTerm
. NN O O
'' NN O O

The NN O O
model NN O I-Generic
acts NN O O
as NN O O
an NN O O
interlingua NN O I-OtherScientificTerm
within NN O O
a NN O O
new NN O O
multi-pathway NN O I-Method
MT NN O I-Method
architecture NN O I-Method
design NN O I-Method
that NN O O
also NN O O
incorporates NN O O
transfer NN O I-OtherScientificTerm
and NN O O
direct NN O I-Generic
approaches NN O I-Generic
into NN O O
a NN O O
single NN O O
system NN O I-Generic
. NN O O

-DOCSTART- NN O O

An NN O O
extension NN O I-Generic
to NN O O
the NN O O
GPSG NN O I-Method
grammatical NN O I-Method
formalism NN O I-Method
is NN O O
proposed NN O O
, NN O O
allowing NN O O
non-terminals NN O I-OtherScientificTerm
to NN O O
consist NN O O
of NN O O
finite NN O O
sequences NN O O
of NN O O
category NN O O
labels NN O O
, NN O O
and NN O O
allowing NN O O
schematic NN O I-OtherScientificTerm
variables NN O I-OtherScientificTerm
to NN O O
range NN O O
over NN O O
such NN O O
sequences NN O O
. NN O O

The NN O O
extension NN O I-Generic
is NN O O
shown NN O O
to NN O O
be NN O O
sufficient NN O O
to NN O O
provide NN O O
a NN O O
strongly NN O O
adequate NN O O
grammar NN O I-Method
for NN O O
crossed NN O I-OtherScientificTerm
serial NN O I-OtherScientificTerm
dependencies NN O I-OtherScientificTerm
, NN O O
as NN O O
found NN O O
in NN O O
e.g. NN O O
Dutch NN O I-OtherScientificTerm
subordinate NN O I-OtherScientificTerm
clauses NN O I-OtherScientificTerm
. NN O O

The NN O O
structures NN O O
induced NN O O
for NN O O
such NN O O
constructions NN O I-Task
are NN O O
argued NN O O
to NN O O
be NN O O
more NN O O
appropriate NN O O
to NN O O
data NN O O
involving NN O O
conjunction NN O O
than NN O O
some NN O O
previous NN O O
proposals NN O O
have NN O O
been NN O O
. NN O O

The NN O O
extension NN O I-Generic
is NN O O
shown NN O O
to NN O O
be NN O O
parseable NN O O
by NN O O
a NN O O
simple NN O O
extension NN O I-Generic
to NN O O
an NN O O
existing NN O O
parsing NN O I-Method
method NN O I-Method
for NN O O
GPSG NN O I-Method
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
an NN O O
approach NN O I-Generic
to NN O O
localizing NN O I-Task
functional NN O I-Task
objects NN O I-Task
in NN O O
surveillance NN O I-Material
videos NN O I-Material
without NN O O
domain NN O I-OtherScientificTerm
knowledge NN O I-OtherScientificTerm
about NN O O
semantic NN O I-OtherScientificTerm
object NN O I-OtherScientificTerm
classes NN O I-OtherScientificTerm
that NN O O
may NN O O
appear NN O O
in NN O O
the NN O O
scene NN O O
. NN O O

Functional NN O I-OtherScientificTerm
objects NN O I-OtherScientificTerm
do NN O O
not NN O O
have NN O O
discriminative NN O I-OtherScientificTerm
appearance NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
shape NN O I-OtherScientificTerm
, NN O O
but NN O O
they NN O I-Generic
affect NN O O
behavior NN O O
of NN O O
people NN O O
in NN O O
the NN O O
scene NN O O
. NN O O

For NN O O
example NN O O
, NN O O
they NN O O
'' NN O O
attract NN O O
'' NN O O
people NN O O
to NN O O
approach NN O O
them NN O O
for NN O O
satisfying NN O O
certain NN O O
needs NN O O
-LRB- NN O O
e.g. NN O O
, NN O O
vending NN O O
machines NN O O
could NN O O
quench NN O O
thirst NN O O
-RRB- NN O O
, NN O O
or NN O O
'' NN O O
repel NN O O
'' NN O O
people NN O O
to NN O O
avoid NN O O
them NN O O
-LRB- NN O O
e.g. NN O O
, NN O O
grass NN O O
lawns NN O O
-RRB- NN O O
. NN O O

Therefore NN O O
, NN O O
functional NN O I-OtherScientificTerm
objects NN O I-OtherScientificTerm
can NN O O
be NN O O
viewed NN O O
as NN O O
'' NN O O
dark NN O O
matter NN O O
'' NN O O
, NN O O
emanating NN O O
'' NN O O
dark NN O O
energy NN O O
'' NN O O
that NN O O
affects NN O O
people NN O O
's NN O O
trajectories NN O O
in NN O O
the NN O O
video NN O O
. NN O O

To NN O O
detect NN O O
'' NN O O
dark NN O O
matter NN O O
'' NN O O
and NN O O
infer NN O O
their NN O O
'' NN O O
dark NN O O
energy NN O O
'' NN O O
field NN O O
, NN O O
we NN O O
extend NN O O
the NN O O
La-grangian NN O I-Method
mechanics NN O I-Method
. NN O O

People NN O O
are NN O O
treated NN O O
as NN O O
particle-agents NN O O
with NN O O
latent NN O O
intents NN O O
to NN O O
approach NN O O
'' NN O O
dark NN O O
matter NN O O
'' NN O O
and NN O O
thus NN O O
satisfy NN O O
their NN O O
needs NN O O
, NN O O
where NN O O
their NN O O
motions NN O O
are NN O O
subject NN O O
to NN O O
a NN O O
composite NN O O
'' NN O O
dark NN O O
energy NN O O
'' NN O O
field NN O O
of NN O O
all NN O O
functional NN O I-OtherScientificTerm
objects NN O I-OtherScientificTerm
in NN O O
the NN O O
scene NN O O
. NN O O

We NN O O
make NN O O
the NN O O
assumption NN O O
that NN O O
people NN O O
take NN O O
globally NN O O
optimal NN O O
paths NN O O
toward NN O O
the NN O O
intended NN O O
'' NN O O
dark NN O O
matter NN O O
'' NN O O
while NN O O
avoiding NN O O
latent NN O O
obstacles NN O O
. NN O O

A NN O O
Bayesian NN O I-Method
framework NN O I-Method
is NN O O
used NN O O
to NN O O
probabilistically NN O O
model NN O O
: NN O O
people NN O I-OtherScientificTerm
's NN O I-OtherScientificTerm
trajectories NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
intents NN O I-OtherScientificTerm
, NN O O
constraint NN O I-OtherScientificTerm
map NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
scene NN O I-OtherScientificTerm
, NN O O
and NN O O
locations NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
functional NN O I-OtherScientificTerm
objects NN O I-OtherScientificTerm
. NN O O

A NN O O
data-driven NN O I-Method
Markov NN O I-Method
Chain NN O I-Method
Monte NN O I-Method
Carlo NN O I-Method
-LRB- NN O I-Method
MCMC NN O I-Method
-RRB- NN O I-Method
process NN O I-Method
is NN O O
used NN O O
for NN O O
inference NN O I-Task
. NN O O

Our NN O O
evaluation NN O O
on NN O O
videos NN O I-Material
of NN O I-Material
public NN O I-Material
squares NN O I-Material
and NN O I-Material
courtyards NN O I-Material
demonstrates NN O O
our NN O O
effectiveness NN O O
in NN O O
localizing NN O I-Task
functional NN O I-Task
objects NN O I-Task
and NN O O
predicting NN O I-Task
people NN O I-Task
's NN O I-Task
trajectories NN O I-Task
in NN O O
unobserved NN O O
parts NN O O
of NN O O
the NN O O
video NN O I-OtherScientificTerm
footage NN O I-OtherScientificTerm
. NN O O

-DOCSTART- NN O O

We NN O O
propose NN O O
a NN O O
process NN O I-Method
model NN O I-Method
for NN O O
hierarchical NN O I-Task
perceptual NN O I-Task
sound NN O I-Task
organization NN O I-Task
, NN O O
which NN O O
recognizes NN O O
perceptual NN O I-OtherScientificTerm
sounds NN O I-OtherScientificTerm
included NN O O
in NN O O
incoming NN O I-OtherScientificTerm
sound NN O I-OtherScientificTerm
signals NN O I-OtherScientificTerm
. NN O O

We NN O O
consider NN O O
perceptual NN O I-Task
sound NN O I-Task
organization NN O I-Task
as NN O O
a NN O O
scene NN O I-Task
analysis NN O I-Task
problem NN O I-Task
in NN O O
the NN O O
auditory NN O I-Material
domain NN O I-Material
. NN O O

Our NN O O
model NN O I-Generic
consists NN O O
of NN O O
multiple NN O O
processing NN O I-Method
modules NN O I-Method
and NN O O
a NN O O
hypothesis NN O I-Method
network NN O I-Method
for NN O O
quantitative NN O O
integration NN O O
of NN O O
multiple NN O O
sources NN O O
of NN O O
information NN O O
. NN O O

When NN O O
input NN O O
information NN O O
for NN O O
each NN O O
processing NN O I-Method
module NN O I-Method
is NN O O
available NN O O
, NN O O
the NN O O
module NN O I-Generic
rises NN O O
to NN O O
process NN O O
it NN O O
and NN O O
asynchronously NN O O
writes NN O O
output NN O O
information NN O O
to NN O O
the NN O O
hypothesis NN O I-Method
network NN O I-Method
. NN O O

On NN O O
the NN O O
hypothesis NN O I-Method
network NN O I-Method
, NN O O
individual NN O O
information NN O O
is NN O O
integrated NN O O
and NN O O
an NN O O
optimal NN O O
internal NN O I-Method
model NN O I-Method
of NN O O
perceptual NN O I-Material
sounds NN O I-Material
is NN O O
automatically NN O O
constructed NN O O
. NN O O

Based NN O O
on NN O O
the NN O O
model NN O I-Method
, NN O O
a NN O O
music NN O I-Method
scene NN O I-Method
analysis NN O I-Method
system NN O I-Method
has NN O O
been NN O O
developed NN O O
for NN O O
acoustic NN O I-OtherScientificTerm
signals NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
ensemble NN O I-OtherScientificTerm
music NN O I-OtherScientificTerm
, NN O O
which NN O O
recognizes NN O O
rhythm NN O I-Material
, NN O O
chords NN O I-Material
, NN O O
and NN O O
source-separated NN O I-Material
musical NN O I-Material
notes NN O I-Material
. NN O O

Experimental NN O O
results NN O O
show NN O O
that NN O O
our NN O O
method NN O I-Generic
has NN O O
permitted NN O O
autonomous NN O O
, NN O O
stable NN O O
and NN O O
effective NN O O
information NN O I-OtherScientificTerm
integration NN O I-OtherScientificTerm
to NN O O
construct NN O O
the NN O O
internal NN O I-Method
model NN O I-Method
of NN O O
hierarchical NN O I-Material
perceptual NN O I-Material
sounds NN O I-Material
. NN O O

-DOCSTART- NN O O

We NN O O
directly NN O O
investigate NN O O
a NN O O
subject NN O O
of NN O O
much NN O O
recent NN O O
debate NN O O
: NN O O
do NN O O
word NN O I-Method
sense NN O I-Method
disambigation NN O I-Method
models NN O I-Method
help NN O O
statistical NN O I-Metric
machine NN O I-Metric
translation NN O I-Metric
quality NN O I-Metric
? NN O O

We NN O O
present NN O O
empirical NN O O
results NN O O
casting NN O O
doubt NN O O
on NN O O
this NN O O
common NN O O
, NN O O
but NN O O
unproved NN O O
, NN O O
assumption NN O O
. NN O O

Using NN O O
a NN O O
state-of-the-art NN O O
Chinese NN O I-Method
word NN O I-Method
sense NN O I-Method
disambiguation NN O I-Method
model NN O I-Method
to NN O O
choose NN O O
translation NN O I-OtherScientificTerm
candidates NN O I-OtherScientificTerm
for NN O O
a NN O O
typical NN O O
IBM NN O I-Method
statistical NN O I-Method
MT NN O I-Method
system NN O I-Method
, NN O O
we NN O O
find NN O O
that NN O O
word NN O I-Method
sense NN O I-Method
disambiguation NN O I-Method
does NN O O
not NN O O
yield NN O O
significantly NN O O
better NN O O
translation NN O I-Metric
quality NN O I-Metric
than NN O O
the NN O O
statistical NN O I-Method
machine NN O I-Method
translation NN O I-Method
system NN O I-Method
alone NN O O
. NN O O

Error NN O I-Method
analysis NN O I-Method
suggests NN O O
several NN O O
key NN O O
factors NN O O
behind NN O O
this NN O O
surprising NN O O
finding NN O O
, NN O O
including NN O O
inherent NN O O
limitations NN O O
of NN O O
current NN O O
statistical NN O I-Method
MT NN O I-Method
architectures NN O I-Method
. NN O O

-DOCSTART- NN O O

Image NN O I-Method
sequence NN O I-Method
processing NN O I-Method
techniques NN O I-Method
are NN O O
used NN O O
to NN O O
study NN O O
exchange NN O I-Task
, NN O I-Task
growth NN O I-Task
, NN O I-Task
and NN O I-Task
transport NN O I-Task
processes NN O I-Task
and NN O O
to NN O O
tackle NN O O
key NN O O
questions NN O O
in NN O O
environmental NN O I-Material
physics NN O I-Material
and NN O O
biology NN O I-Material
. NN O O

These NN O O
applications NN O I-Generic
require NN O O
high NN O O
accuracy NN O I-Metric
for NN O O
the NN O O
estimation NN O I-Task
of NN O I-Task
the NN O I-Task
motion NN O I-Task
field NN O I-Task
since NN O O
the NN O O
most NN O O
interesting NN O O
parameters NN O O
of NN O O
the NN O O
dynamical NN O I-Method
processes NN O I-Method
studied NN O O
are NN O O
contained NN O O
in NN O O
first-order NN O I-OtherScientificTerm
derivatives NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
motion NN O I-OtherScientificTerm
field NN O I-OtherScientificTerm
or NN O O
in NN O O
dynamical NN O I-OtherScientificTerm
changes NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
moving NN O I-OtherScientificTerm
objects NN O I-OtherScientificTerm
. NN O O

Therefore NN O O
the NN O O
performance NN O O
and NN O O
optimization NN O I-Task
of NN O I-Task
low-level NN O I-Task
motion NN O I-Task
estimators NN O I-Task
is NN O O
discussed NN O O
. NN O O

A NN O O
tensor NN O I-Method
method NN O I-Method
tuned NN O O
with NN O O
carefully NN O O
optimized NN O O
derivative NN O I-Method
filters NN O I-Method
yields NN O O
reliable NN O O
and NN O O
dense NN O O
displacement NN O I-OtherScientificTerm
vector NN O I-OtherScientificTerm
fields NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
DVF NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
with NN O O
an NN O O
accuracy NN O I-Metric
of NN O O
up NN O O
to NN O O
a NN O O
few NN O O
hundredth NN O O
pixels/frame NN O I-Metric
for NN O O
real-world NN O I-Material
images NN O I-Material
. NN O O

The NN O O
accuracy NN O I-Metric
of NN O O
the NN O O
tensor NN O I-Method
method NN O I-Method
is NN O O
verified NN O O
with NN O O
computer-generated NN O I-Material
sequences NN O I-Material
and NN O O
a NN O O
calibrated NN O I-Material
image NN O I-Material
sequence NN O I-Material
. NN O O

With NN O O
the NN O O
improvements NN O O
in NN O O
accuracy NN O I-Metric
the NN O O
motion NN O I-Task
estimation NN O I-Task
is NN O O
now NN O O
rather NN O O
limited NN O O
by NN O O
imperfections NN O O
in NN O O
the NN O O
CCD NN O I-OtherScientificTerm
sensors NN O I-OtherScientificTerm
, NN O O
especially NN O O
the NN O O
spatial NN O I-OtherScientificTerm
nonuni-formity NN O I-OtherScientificTerm
in NN O O
the NN O O
responsivity NN O I-OtherScientificTerm
. NN O O

With NN O O
a NN O O
simple NN O O
two-point NN O I-Method
calibration NN O I-Method
, NN O O
these NN O O
effects NN O O
can NN O O
efficiently NN O O
be NN O O
suppressed NN O O
. NN O O

The NN O O
application NN O O
of NN O O
the NN O O
techniques NN O I-Generic
to NN O O
the NN O O
analysis NN O I-Task
of NN O I-Task
plant NN O I-Task
growth NN O I-Task
, NN O O
to NN O O
ocean NN O I-Task
surface NN O I-Task
microturbulence NN O I-Task
in NN O I-Task
IR NN O I-Task
image NN O I-Task
sequences NN O I-Task
, NN O O
and NN O O
to NN O O
sediment NN O I-Task
transport NN O I-Task
is NN O O
demonstrated NN O O
. NN O O

-DOCSTART- NN O O

We NN O O
present NN O O
a NN O O
Czech-English NN O I-Method
statistical NN O I-Method
machine NN O I-Method
translation NN O I-Method
system NN O I-Method
which NN O O
performs NN O O
tree-to-tree NN O I-Task
translation NN O I-Task
of NN O I-Task
dependency NN O I-Task
structures NN O I-Task
. NN O O

The NN O O
only NN O O
bilingual NN O I-Material
resource NN O I-Material
required NN O O
is NN O O
a NN O O
sentence-aligned NN O I-Material
parallel NN O I-Material
corpus NN O I-Material
. NN O O

All NN O O
other NN O O
resources NN O O
are NN O O
monolingual NN O O
. NN O O

We NN O O
also NN O O
refer NN O O
to NN O O
an NN O O
evaluation NN O O
method NN O O
and NN O O
plan NN O O
to NN O O
compare NN O O
our NN O O
system NN O I-Generic
's NN O O
output NN O O
with NN O O
a NN O O
benchmark NN O I-Generic
system NN O I-Generic
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
describes NN O O
the NN O O
understanding NN O O
process NN O O
of NN O O
the NN O O
spatial NN O I-OtherScientificTerm
descriptions NN O I-OtherScientificTerm
in NN O O
Japanese NN O I-Material
. NN O O

In NN O O
order NN O O
to NN O O
understand NN O O
the NN O O
described NN O O
world NN O O
, NN O O
the NN O O
authors NN O O
try NN O O
to NN O O
reconstruct NN O O
the NN O O
geometric NN O I-Method
model NN O I-Method
of NN O O
the NN O O
global NN O O
scene NN O O
from NN O O
the NN O O
scenic NN O O
descriptions NN O O
drawing NN O O
a NN O O
space NN O O
. NN O O

It NN O O
is NN O O
done NN O O
by NN O O
an NN O O
experimental NN O O
computer NN O I-Method
program NN O I-Method
SPRINT NN O I-Method
, NN O O
which NN O O
takes NN O O
natural NN O I-Material
language NN O I-Material
texts NN O I-Material
and NN O O
produces NN O O
a NN O O
model NN O I-Generic
of NN O O
the NN O O
described NN O O
world NN O O
. NN O O

To NN O O
reconstruct NN O O
the NN O O
model NN O I-Generic
, NN O O
the NN O O
authors NN O O
extract NN O O
the NN O O
qualitative NN O I-OtherScientificTerm
spatial NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
from NN O O
the NN O O
text NN O O
, NN O O
and NN O O
represent NN O O
them NN O O
as NN O O
the NN O O
numerical NN O I-OtherScientificTerm
constraints NN O I-OtherScientificTerm
on NN O O
the NN O O
spatial NN O I-OtherScientificTerm
attributes NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
entities NN O I-OtherScientificTerm
. NN O O

This NN O O
makes NN O O
it NN O O
possible NN O O
to NN O O
express NN O O
the NN O O
vagueness NN O O
of NN O O
the NN O O
spatial NN O I-OtherScientificTerm
concepts NN O I-OtherScientificTerm
and NN O O
to NN O O
derive NN O O
the NN O O
maximally NN O O
plausible NN O O
interpretation NN O O
from NN O O
a NN O O
chunk NN O O
of NN O O
information NN O O
accumulated NN O O
as NN O O
the NN O O
constraints NN O O
. NN O O

The NN O O
interpretation NN O O
reflects NN O O
the NN O O
temporary NN O O
belief NN O O
about NN O O
the NN O O
world NN O O
. NN O O

-DOCSTART- NN O O

Learning NN O I-Task
video NN O I-Task
representation NN O I-Task
is NN O O
not NN O O
a NN O O
trivial NN O O
task NN O O
, NN O O
as NN O O
video NN O O
is NN O O
an NN O O
information-intensive NN O O
media NN O O
where NN O O
each NN O O
frame NN O O
does NN O O
not NN O O
exist NN O O
independently NN O O
. NN O O

Locally NN O O
, NN O O
a NN O O
video NN O O
frame NN O O
is NN O O
visually NN O O
and NN O O
semantically NN O O
similar NN O O
with NN O O
its NN O O
adjacent NN O O
frames NN O O
. NN O O

Holistically NN O O
, NN O O
a NN O O
video NN O O
has NN O O
its NN O O
inherent NN O O
structure NN O O
-- NN O O
the NN O O
correlations NN O O
among NN O O
video NN O O
frames NN O O
. NN O O

For NN O O
example NN O O
, NN O O
even NN O O
the NN O O
frames NN O O
far NN O O
from NN O O
each NN O O
other NN O O
may NN O O
also NN O O
hold NN O O
similar NN O O
semantics NN O I-OtherScientificTerm
. NN O O

Such NN O O
context NN O I-OtherScientificTerm
information NN O I-OtherScientificTerm
is NN O O
therefore NN O O
important NN O O
to NN O O
characterize NN O O
the NN O O
intrinsic NN O I-Method
representation NN O I-Method
of NN O I-Method
a NN O I-Method
video NN O I-Method
frame NN O I-Method
. NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
present NN O O
a NN O O
novel NN O O
approach NN O I-Generic
to NN O O
learn NN O O
the NN O O
deep NN O I-Method
video NN O I-Method
representation NN O I-Method
by NN O O
exploring NN O O
both NN O O
local NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
holistic NN O I-OtherScientificTerm
contexts NN O I-OtherScientificTerm
. NN O O

Specifically NN O O
, NN O O
we NN O O
propose NN O O
a NN O O
triplet NN O I-Method
sampling NN O I-Method
mechanism NN O I-Method
to NN O O
encode NN O O
the NN O O
local NN O I-OtherScientificTerm
temporal NN O I-OtherScientificTerm
relationship NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
adjacent NN O I-OtherScientificTerm
frames NN O I-OtherScientificTerm
based NN O O
on NN O O
their NN O O
deep NN O I-Method
representations NN O I-Method
. NN O O

In NN O O
addition NN O O
, NN O O
we NN O O
incorporate NN O O
the NN O O
graph NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
the NN O I-OtherScientificTerm
video NN O I-OtherScientificTerm
, NN O O
as NN O O
a NN O O
priori NN O I-OtherScientificTerm
, NN O O
to NN O O
holistically NN O O
preserve NN O O
the NN O O
inherent NN O O
correlations NN O O
among NN O O
video NN O O
frames NN O O
. NN O O

Our NN O O
approach NN O I-Generic
is NN O O
fully NN O O
unsupervised NN O O
and NN O O
trained NN O O
in NN O O
an NN O O
end-to-end NN O I-Method
deep NN O I-Method
convolutional NN O I-Method
neu-ral NN O I-Method
network NN O I-Method
architecture NN O I-Method
. NN O O

By NN O O
extensive NN O O
experiments NN O O
, NN O O
we NN O O
show NN O O
that NN O O
our NN O O
learned NN O I-Generic
representation NN O I-Generic
can NN O O
significantly NN O O
boost NN O O
several NN O O
video NN O I-Task
recognition NN O I-Task
tasks NN O I-Task
-LRB- NN O O
retrieval NN O I-Task
, NN O O
classification NN O I-Task
, NN O O
and NN O O
highlight NN O I-Task
detection NN O I-Task
-RRB- NN O O
over NN O O
traditional NN O O
video NN O I-Generic
representations NN O I-Generic
. NN O O

-DOCSTART- NN O O

For NN O O
mobile NN O I-Task
speech NN O I-Task
application NN O I-Task
, NN O O
speaker NN O I-Metric
DOA NN O I-Metric
estimation NN O I-Metric
accuracy NN O I-Metric
, NN O O
interference NN O I-Metric
robustness NN O I-Metric
and NN O O
compact NN O I-Metric
physical NN O I-Metric
size NN O I-Metric
are NN O O
three NN O O
key NN O O
factors NN O O
. NN O O

Considering NN O O
the NN O O
size NN O O
, NN O O
we NN O O
utilized NN O O
acoustic NN O I-Method
vector NN O I-Method
sensor NN O I-Method
-LRB- NN O I-Method
AVS NN O I-Method
-RRB- NN O I-Method
and NN O O
proposed NN O O
a NN O O
DOA NN O I-Method
estimation NN O I-Method
algorithm NN O I-Method
previously NN O O
-LSB- NN O O
1 NN O O
-RSB- NN O O
, NN O O
offering NN O O
high NN O O
accuracy NN O O
with NN O O
larger-than-15dB NN O O
SNR NN O O
but NN O O
is NN O O
deteriorated NN O O
by NN O O
non-speech NN O I-OtherScientificTerm
interferences NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
NSI NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
. NN O O

This NN O O
paper NN O O
develops NN O O
a NN O O
robust NN O I-Method
speaker NN O I-Method
DOA NN O I-Method
estimation NN O I-Method
algorithm NN O I-Method
. NN O O

It NN O I-Generic
is NN O O
achieved NN O O
by NN O O
deriving NN O O
the NN O O
inter-sensor NN O I-Method
data NN O I-Method
ratio NN O I-Method
model NN O I-Method
of NN O O
an NN O O
AVS NN O I-Method
in NN O O
bispectrum NN O I-Method
domain NN O I-Method
-LRB- NN O I-Method
BISDR NN O I-Method
-RRB- NN O I-Method
and NN O O
exploring NN O O
the NN O O
favorable NN O I-Generic
properties NN O I-Generic
of NN O O
bispectrum NN O I-Method
, NN O O
such NN O O
as NN O O
zero NN O I-OtherScientificTerm
value NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
Gaussian NN O I-OtherScientificTerm
process NN O I-OtherScientificTerm
and NN O O
different NN O O
distribution NN O I-OtherScientificTerm
of NN O I-OtherScientificTerm
speech NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
NSI NN O B-OtherScientificTerm
. NN O O

Specifically NN O O
, NN O O
a NN O O
reliable NN O O
bispectrum NN O I-Method
mask NN O I-Method
is NN O O
generated NN O O
to NN O O
guarantee NN O O
that NN O O
the NN O O
speaker NN O I-OtherScientificTerm
DOA NN O I-OtherScientificTerm
cues NN O I-OtherScientificTerm
, NN O O
derived NN O O
from NN O O
BISDR NN O I-Method
, NN O O
are NN O O
robust NN O O
to NN O O
NSI NN O I-OtherScientificTerm
in NN O O
terms NN O O
of NN O O
speech NN O I-OtherScientificTerm
sparsity NN O I-OtherScientificTerm
and NN O O
large NN O O
bispectrum NN O I-OtherScientificTerm
amplitude NN O I-OtherScientificTerm
of NN O O
the NN O O
captured NN O O
signals NN O O
. NN O O

Intensive NN O O
experiments NN O O
demonstrate NN O O
an NN O O
improved NN O O
performance NN O O
of NN O O
our NN O O
proposed NN O O
algorithm NN O I-Generic
under NN O O
various NN O O
NSI NN O I-OtherScientificTerm
conditions NN O I-OtherScientificTerm
even NN O O
when NN O O
SIR NN O I-Metric
is NN O O
smaller NN O O
than NN O O
0dB NN O O
. NN O O

-DOCSTART- NN O O

In NN O O
this NN O O
paper NN O O
, NN O O
we NN O O
want NN O O
to NN O O
show NN O O
how NN O O
the NN O O
morphological NN O I-Method
component NN O I-Method
of NN O O
an NN O O
existing NN O O
NLP-system NN O I-Method
for NN O I-Method
Dutch NN O I-Method
-LRB- NN O I-Method
Dutch NN O I-Method
Medical NN O I-Method
Language NN O I-Method
Processor NN O I-Method
- NN O I-Method
DMLP NN O I-Method
-RRB- NN O I-Method
has NN O O
been NN O O
extended NN O O
in NN O O
order NN O O
to NN O O
produce NN O O
output NN O O
that NN O O
is NN O O
compatible NN O O
with NN O O
the NN O O
language NN O I-Method
independent NN O I-Method
modules NN O I-Method
of NN O O
the NN O O
LSP-MLP NN O I-Method
system NN O I-Method
-LRB- NN O I-Method
Linguistic NN O I-Method
String NN O I-Method
Project NN O I-Method
- NN O I-Method
Medical NN O I-Method
Language NN O I-Method
Processor NN O I-Method
-RRB- NN O I-Method
of NN O O
the NN O O
New NN O O
York NN O O
University NN O O
. NN O O

The NN O O
former NN O I-Generic
can NN O O
take NN O O
advantage NN O O
of NN O O
the NN O O
language NN O O
independent NN O O
developments NN O O
of NN O O
the NN O O
latter NN O I-Generic
, NN O O
while NN O O
focusing NN O O
on NN O O
idiosyncrasies NN O I-OtherScientificTerm
for NN O O
Dutch NN O I-Material
. NN O O

This NN O O
general NN O O
strategy NN O O
will NN O O
be NN O O
illustrated NN O O
by NN O O
a NN O O
practical NN O O
application NN O I-Generic
, NN O O
namely NN O O
the NN O O
highlighting NN O I-Task
of NN O I-Task
relevant NN O I-Generic
information NN O I-Generic
in NN O O
a NN O O
patient NN O I-OtherScientificTerm
discharge NN O I-OtherScientificTerm
summary NN O I-OtherScientificTerm
-LRB- NN O I-OtherScientificTerm
PDS NN O I-OtherScientificTerm
-RRB- NN O I-OtherScientificTerm
by NN O O
means NN O O
of NN O O
modern NN O O
HyperText NN O I-Method
Mark-Up NN O I-Method
Language NN O I-Method
-LRB- NN O I-Method
HTML NN O I-Method
-RRB- NN O I-Method
technology NN O I-Method
. NN O O

Such NN O O
an NN O O
application NN O I-Generic
can NN O O
be NN O O
of NN O O
use NN O O
for NN O O
medical NN O I-Task
administrative NN O I-Task
purposes NN O I-Task
in NN O O
a NN O O
hospital NN O O
environment NN O O
. NN O O

-DOCSTART- NN O O

CriterionSM NN O I-Task
Online NN O I-Task
Essay NN O I-Task
Evaluation NN O I-Task
Service NN O I-Task
includes NN O O
a NN O O
capability NN O O
that NN O O
labels NN O O
sentences NN O O
in NN O O
student NN O O
writing NN O O
with NN O O
essay-based NN O I-OtherScientificTerm
discourse NN O I-OtherScientificTerm
elements NN O I-OtherScientificTerm
-LRB- NN O O
e.g. NN O O
, NN O O
thesis NN O I-OtherScientificTerm
statements NN O I-OtherScientificTerm
-RRB- NN O O
. NN O O

We NN O O
describe NN O O
a NN O O
new NN O O
system NN O I-Generic
that NN O O
enhances NN O O
Criterion NN O I-OtherScientificTerm
's NN O I-OtherScientificTerm
capability NN O I-OtherScientificTerm
, NN O O
by NN O O
evaluating NN O O
multiple NN O O
aspects NN O O
of NN O O
coherence NN O I-Metric
in NN O I-Metric
essays NN O I-Metric
. NN O O

This NN O O
system NN O I-Generic
identifies NN O O
features NN O I-OtherScientificTerm
of NN O O
sentences NN O O
based NN O O
on NN O O
semantic NN O I-Metric
similarity NN O I-Metric
measures NN O I-Metric
and NN O O
discourse NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
. NN O O

A NN O O
support NN O I-Method
vector NN O I-Method
machine NN O I-Method
uses NN O O
these NN O O
features NN O I-OtherScientificTerm
to NN O O
capture NN O O
breakdowns NN O I-OtherScientificTerm
in NN O I-OtherScientificTerm
coherence NN O I-OtherScientificTerm
due NN O O
to NN O O
relatedness NN O O
to NN O O
the NN O O
essay NN O O
question NN O O
and NN O O
relatedness NN O O
between NN O O
discourse NN O I-OtherScientificTerm
elements NN O I-OtherScientificTerm
. NN O O

Intra-sentential NN O I-Metric
quality NN O I-Metric
is NN O O
evaluated NN O O
with NN O O
rule-based NN O I-Method
heuristics NN O I-Method
. NN O O

Results NN O O
indicate NN O O
that NN O O
the NN O O
system NN O I-Generic
yields NN O O
higher NN O O
performance NN O O
than NN O O
a NN O O
baseline NN O I-Generic
on NN O O
all NN O O
three NN O O
aspects NN O O
. NN O O

-DOCSTART- NN O O

This NN O O
paper NN O O
presents NN O O
an NN O O
algorithm NN O I-Generic
for NN O O
labeling NN O I-Task
curvilinear NN O I-Task
structure NN O I-Task
at NN O O
multiple NN O O
scales NN O O
in NN O O
line NN O I-Material
drawings NN O I-Material
and NN O O
edge NN O I-Material
images NN O I-Material
Symbolic NN O O
CURVE-ELEMENT NN O I-OtherScientificTerm
tokens NN O I-OtherScientificTerm
residing NN O O
in NN O O
a NN O O
spatially-indexed NN O I-OtherScientificTerm
and NN O I-OtherScientificTerm
scale-indexed NN O I-OtherScientificTerm
data NN O I-OtherScientificTerm
structure NN O I-OtherScientificTerm
denote NN O O
circular NN O O
arcs NN O O
fit NN O O
to NN O O
image NN O I-Material
data NN O I-Material
. NN O O

Tokens NN O O
are NN O O
computed NN O O
via NN O O
a NN O O
small-to-large NN O I-Method
scale NN O I-Method
grouping NN O I-Method
procedure NN O I-Method
employing NN O O
a NN O O
'' NN O O
greedy NN O O
'' NN O O
, NN O O
best-first NN O O
, NN O O
strategy NN O O
for NN O O
choosing NN O O
the NN O O
support NN O O
of NN O O
new NN O O
tokens NN O O
. NN O O

The NN O O
resulting NN O O
image NN O I-Task
description NN O I-Task
is NN O O
rich NN O O
and NN O O
redundant NN O O
in NN O O
that NN O O
a NN O O
given NN O O
segment NN O O
of NN O O
image NN O I-OtherScientificTerm
contour NN O I-OtherScientificTerm
may NN O O
be NN O O
described NN O O
by NN O O
multiple NN O O
tokens NN O O
at NN O O
different NN O O
scales NN O O
, NN O O
and NN O O
by NN O O
more NN O O
than NN O O
one NN O O
token NN O O
at NN O O
any NN O O
given NN O O
scale NN O O
. NN O O

This NN O O
property NN O O
facilitates NN O O
selection NN O O
and NN O O
characterization NN O O
of NN O O
portions NN O O
of NN O O
the NN O O
image NN O O
based NN O O
on NN O O
local NN O I-OtherScientificTerm
CURVE-ELEMENT NN O I-OtherScientificTerm
attributes NN O I-OtherScientificTerm
. NN O O

