►
Description
Could debates push forward collective understanding? If possible, how do we make it happen? The analysis of large datasets documenting a century of scientific debates through criticizing and endorsing citations between millions of scholars and two decades of edit battles between Wikipedia contributors provides insights into the social mechanisms of debates and informs institutional arrangements to facilitate knowledge creation from debates on social media.
Date: 3/5/20
Presenter: Lingfei Wu
Institution: University of Pittsburgh
A
All
right
so
Oh
to
everyone
there,
if
you
could
type
in
your,
could
you
go
in
and
change
your
name
on
your
screen
so
that
it
reflects
your
real
name,
just
search
of
tracking
purposes
for
MC
and
for
those
of
you
who
absolutely
have
to
remain
anonymous?
Please
ignore
that
request.
I
know
there
are
some
of
you
like
that.
A
Okay
with
that
being
said,
it
is
my
great
pleasure
to
introduce
you
to
our
speaker
today,
link'
womb,
which,
from
the
University
of
Pittsburgh
and
their
School
of
Computing
and
information.
His
work
is
distinct
from
a
lot
of
the
work
live
here
so
far
because
he's
interested
in
the
geometry
of
thinking.
He
himself
is
a
computational
scientist
and
he
applies
this
approach
to
the
geometry
of
thinking
to
lots
of
topics,
and
one
of
those
is
of
course,
debates
and
looking
at
how
expert
opinion
kind
of
flows
within
and
among
the
groups
at
this
point.
B
B
This
is
always
true,
so
one
in
the
research,
even
in
the
level
side
of
the
panel
of
the
figure,
you
can
see
the
distribution
of
how
polarized
these
weekly
users
are
by
measuring
the
engagement,
a
fixed
set
of
either
liberal
or
conservative
wiki
pages.
So
you
can,
if
you
are
contributing
to
a
more
liberal,
wiki
page,
then
we'll
put
you
in
some
kind
of
in
some
position
in
this
spectrum,
and
it
will
appeal
to
more
conservative
page
will
also
do
that.
So
this
is
a
from
this.
B
You
can
map
out
the
whole
political
tenderness
or
preference
tendencies
of
these
wiki
contributors
and
from
there
you
can
see.
Indeed,
people
have
having
very
different
ideas,
including
all
the
way
to
the
left,
which
is
like
highly
liberal
and
already
to
the
right,
which
is
highly
conservative,
but
also
with
a
majority
of
audience.
All
contributors
lie
in
a
in
the
neutral
area,
which
is
like
near
a
zero
values.
B
So
this
is
like
other
editors
and
contributors
of
the
wikipedia
wiki
page
and
now
the
research
question
is:
if
we
have
a
page
and
as
we
all
know,
we
keep
Wikipedia
is
the
open
platform
and
that
engage
a
lot
of
different
contributions
and
people
can
talk
to
each
other
in
the
comp
age.
People
can
change
each
other's
Eddy's.
People
can
make
Windows
there
all
kinds
of
interactions
going
on.
Allow
you
to
create
and/or
converge
Norwich
in
the
conversation
of
ways
right.
B
So
if
we
have
in
a
page-
and
we
cheat
all
these
people
on
the
contributor
same
page
as
a
same
thing,
and
if
we
have
a
team
that
is
more
polarized
rather
than
less
for
ice,
so
here
like
we,
we
can.
We
can
also
talk
about
like
the
best
attorney.
You
can
call
it
for
ice
or
like
a
fragment
here.
In
this
research
there
not
much
detail
differentiations
between
those
so
base.
We
are
talking
about
in
qualitative
ways.
They
see
like
how
much
the
contributors
of
this
page
are
disputing
on.
B
The
two
ends
left
in
white
ends
of
the
spectrum
rather
than
they
concentrate
on
the
middle
and
how
this
distribution
affect
the
quality
of
the
Wikipedia
page.
So
the
others
from
this
obtained
a
grant
of
data
a
grant
to
the
light
data
coordinating
with
the
wiki
with
the
foundation,
research
teams,
and
they
have
a
machine
learning
model
to
exchange
low
quality
evaluation
standards
from
a
crowd-sourced
grounds
with
data
set
so
that
they
can
apply
that
to
more
than
400,000
wiki
pages
and
give
low
quality
score
from
that.
B
B
You
have
a
different
quality
levels
of
wiki
pages,
ranging
from
all
the
way
from
stop
to
feature
articles,
and
you
also
have
another
independent
variable
which
is
like
how
poor
eyes
or
how
different,
how
much
variance
within
these
in
terms
of
political
preference.
In
terms
of
within
this
group
of
contributors
and
very
interesting,
we
found
that
I
mean
they
found
that,
with
more
prize
crowd,
even
controlling
other
factors
that
you
can
think
of
the
number
of
computers,
the
lens
of
articles
and
the
number
of
edits.
B
All
these
things,
you
still
see
a
very
significant
positive
correlation
between
polarization
and
quality
of
articles.
So
this
this
is,
we
least
point
out
a
very
interesting
possibility
over
there.
These
others
summarized
in
their
article
that
is
polarizing
here
or
is
the
difference
of
opinion
or
respect.
What
is
okay,
what
is
like
in
Wikipedia,
we
have
very
limits
space
like
virtual
space.
You
know
in
the
sense
that
so
this
is
a
definition
of
democracy.
B
You
guys
have
to
figure
out
a
immigration
issue
of
democracy
by
yourself,
so
you
have
highly
liberal
people
have
hanakos
available,
but
you
need
to
discuss
and
formalize
and
socialize
and
fully
oxidized
and
stabilize
your
idea
on
this
page.
Now
in
this
new
Asian,
you
have
people
with
energy.
You
have
people
want
to
convince
each
other.
You
have
people
want
to
contribute,
so
you
have
a
lot
of
like
personal
energy
going
on,
but
you
also
have
a
system
in
which
people
will
coordinate.
We
have
to
come
up
with
a
result.
B
We
have,
they
will
have
different
levels
of
editor
to
coordinating,
and
maybe
maybe
in
this
kind
of
scenarios,
that
debate
is
not
always
there.
We
actually
push
for
the
understanding
of
Norwich
and
the
better
definition
and
the
more
communication
skills
like
making
metaphors
and
this
and
that
used
to
make
the
adequate
self
more
accessible
to
a
general
audience
or
like
make
all
like
every
fu--
node
will
be
carefully
checked
by
people
on
people
with
a
different
point
of
view
of
in
terms
of
critical
spectrum.
B
So
all
these
things,
maybe
the
complex
landscape
of
the
the
team
composition
problem-
will
all
contribute
to
actually
the
consensus
in
Norwich.
So
this
is
a
very
interesting
finding,
as
I
said
in
the
background.
Ok,
so
another
finding
about
another
interesting
literature,
I
want
to
point
your
attention
to
is
from
the
same
thing
of
the
Norwich
11:18
by
James
Evans,
but
published
on
an
American
Journal
of
sociology
called
ambiguity
and
engagement.
So
this
is
even
the
last
research
is
talking
about
whether
the
engagement
will
create
Norwich.
B
Now
this
lease
article
is
looking
at
the
front
things
from
the
other
way.
What
kind
of
Norwich
will
father
the
engagement
of
discussion
so
like
engagement,
knowledge,
they're
kind
of
like
both
of
cause
and
consequence
of
each
other?
But
you
can
you
keep
people
engaged?
We
have
been
engaging
in
creating
knowledge
in
science
for
hundreds
of
years,
okay.
So
this
is
a
very
interesting
article.
Looking
at
two
variables,
one
virgo
is,
would
ambiguity
so
now
the
wording
big
ambiguous.
B
You
can
also
call
it
word
interchangeability
or
what
a
substitution
that
is
to
measure
to
what
extent
a
word
can
be
replaced.
By
other
words,
for
example,
you
can
look
at
this
kind
of
contest.
You
see
like
words
like
mechanical,
it's
used
only
in
a
very
specific
sense
that
it
cannot
be
easily
replaced
by
other
words,
but
were
like,
like
structure
or
other
things
or
like
form
or
arrangement
states.
These
kind
of
words
are
more
easily
more
Pozzuoli,
be
replaced
by
other
words.
B
So
the
measure
is,
if
you
find
a
certain
word
and
you
identify
other
as
a
small
set
of
words
with
similar
meanings
that
you
want
to
observe
or
measure
actually
in
interchangeability
between
these
words
and
then
you
select
all
the
sentence
to
see
to
counter
frequency
at
which
or
in
among
all
these
contents.
Only
one
word
out
word
will
be
use
highly
frequently,
when
other
words
are
less
likely
or
the
probability
of
using
different
words
in
the
same
context
is
almost
uniform,
like
maximizing
of
variance
and
entropy,
so
in
the
other.
B
So
if
the
word
is
using
exchanger
interchangeability
in
the
tent
body
with
our
words,
then
we
call
this
wall.
It's
like
any
girls
word,
because
when
all
the
imagine
is
what
the
author
may
also
measure
other
words
the
use
to
express
the
same
sense,
but
then,
if
this
word
is
only
you
so
fixed
within
this
kind
of
certain
context,
you
cannot
repress
a
the
words
name.
Is
law
is
less
ambiguous,
so
this
is
a
language
ambiguity
and
the
other
on
the
other
hands
they
measured.
B
The
engagement
like
to
what
extent
the
following
citations
to
the
paper
so
further
for
the
for
the
article
research
article.
You
can
already
mention
the
the
average
of
all
language
ambiguity
and
then
now
you
can
measure
to
extend
the
foreign
articles
or
following
others
when
they
cite
is
article
2.
They
also
cite
each
other.
So
a
easy
way
to
understand
these.
If
you
use
very
ambiguous
words
like
you,
like
confusion,
is
talking
about
love
or
Jesus
talking
about
love,
then
all
these
disciples
will
talk
to
each
other
you're
like
what
did
Jesus
jesus
said.
B
What
what
does
he
mean
about
about
love?
Does
the
law?
What
kind
of
context
does
it
is
a
price
they
probably
like
10
different
versions
of
it's?
The
explanation
from
different
disciples
then
like
they
are?
They
have
to
cite
each
other
to
confirm
each
other,
but
this
process
is
also
the
creation
kind
of
engagement.
So
this
will
this
too,
by
measuring
to
extend
the
following
articles
in
the
only
cite
the
original
Amoco
informing
different
branch,
which
is
like
low
entropy
and
the
low
engagement
all
the
way
to
the
rights
paradigm.
That's
all
the
following.
B
Citations
also
cite
each
other
coercing
each
other,
not
only
praising
branch,
but
also
breaking
these
branches
and
cross
custard
and
fragments.
Then
it's
a
high
engagement,
so
you
can
use
these
to
measure
out
these
kinds
of
kind
of
situations
and
turns
out
of
these.
Two
variables
are
highly
correlated.
B
They
are
highly
correlated
what
you
are
using
very
ambiguous
words:
you
create
an
environment
that
allows
competing
interpretations,
come
into
the
conversation
and
you
actually
stimulates
and
motivate
social
learning,
because
people
want
to
know
what
is
happens
and
what
on
what
either
Jesus,
either
Confucius,
either
Max
Weber
just
said
so
they
may
have
to
keep
citing
from
each
other
keeping
the
discussion
going
on.
So
eventually
you
actually
also
true,
very
ironically,
by
talking
about
things
and
statically,
creating
uncertainty.
B
You
actually
also
push
forward
a
knowledge
of
money,
so
that's
a
putting
together
this
versus
previous
research.
We
we
see
a
very
interesting
picture
that
having
people
have
a
different
opinion
with
you.
It's
not
always
a
bad
thing.
Maybe
it
could
be
the
cause
of
knowledge
that
you
will
keep
engage
in
a
debate
and
then,
like
also
another
thing
is:
if,
in
the
in
the
in
the
in
a
production
of
knowledge
itself,
no
converging
too
much
early,
maybe
not
always
a
bad
thing.
People
we
have
different
ideas.
B
People
probably
have
different
competing
interpretations
of
the
same
terms
or
they
use
it
differently.
Then
this
kind
of
thing
is
not
special.
Of
course
it's
you
cannot
write
down
things
very
quickly
on
the
textbook
of
course,
but
then
this
kind
of
situation
also
create
the
oken
environment
for
future
engagement.
So
this
is
like
very
interesting
to
research,
talking
about
the
cause
and
consequence
between
knowledge
and
the
conversation
and
before
a
makhotin.
B
That
is
not
highlights
by
the
authors
in
this
tool,
research,
that
is
the
difference
between
soft
knowledge
and
Honora,
and
I
really
want
to
invite
you
to
see
collectively.
Think
about
this.
This
situation,
that
when
we
talk
about
knowledge,
what
kind
of
knowledge
are
we
talking
about
and
when
we
talk
about
debate
and
push
for
what
understanding
in
what
situation
made
that
happens
more
efficiently,
rather
than
less
efficient,
efficient?
Okay,
so
one
reservation
you
can
draw
from
this
beauty,
beauty
and
engagement
paper.
B
More
ambiguity
used
that
in
humanity
right
and
in
general,
in
hard
science,
empirical
science
in
science,
we
involve
more
equations
measuring
devices,
experiments
data
that
if
we
record
a
hard
science
or
her
knowledge,
then
the
words
are
used.
More
specifically,
this
is
very
understandable
because
when
we
talk
about
Newton's
equations,
when
we
talk
about
gravity,
when
we
talk
about
relativity
theory,
when
we
talk
about
corner
physics,
it's
well
defined,
you
cannot
just
change
the
name
of
gravity
and
talking
about
a
gravity
of
your
I.
B
Don't
know
your
heart
who's
like
a
comfort
food
right,
you
can
use
it
in
a
beauty
contest
you
can,
but
you
cannot
use
the
in
in
in
research,
audible
contacts,
but
in
sociology
in
humanity
it's
less
fixed
people.
Knowledge
is
more
fuel
through
it,
like
you
can
use
words
in
in
different
water
to
explain
the
same,
complex
meaning,
and
it
will
give
you
this
article,
the
first
article
I
mentioned.
B
You
also
see
this
kind
of
difference,
so
if
we
even
will
observe
and
our
chronology
the
the
the
positive
correlation
between
the
polarization
and
quality
then
are
those
correlation,
strong
or
weak,
on
cause
different
kinds
of
knowledge.
Maybe
yes,
we
see,
we
see
the
significant
difference
here
that
this
kind
of
debate
and
discussion
is
more
pushing
for
is
increasing
the
knowledge
quality
of
this
article
quality
of
political
of
politics
and
pseudoscience,
and
the
social
issues
more
rather
than
science.
So
it
looks
like
in
terms
of
political
issues
in
terms
of
social
issues.
B
B
Finding
I
do
from
their
context,
which
I
am
voting
a
little
bit
in
my
explorations
of
Norwich,
and
also
try
to
discuss
that
the
user
as
a
context
of
studying
debating
science.
So
allow
me
to
use
one
more
slides
to
talk
a
little
bit
more
about
self-knowledge
and
our
knowledge
and
what
or
22p
it
measures
of
these
two
kind
of
things,
and
so
we
can
build
up
a
deeper
understanding
and
a
bigger
picture
right
if
I
may
naively
summarize.
Soft
knowledge
is
something
like
contact
space,
it's
very
hard
to
move
across
time
space.
B
What
we
are
talking
about
now,
it's
a
sort
of
knowledge,
can
debate
push
forward
knowledge.
That
is
a
self-knowledge
because
we
I
can't
I
can't
just
show
you
one
equation
like
really
long
and
very
have
the
highly
soon
answered
all
of
you
say
yes
or
not.
We
can
point
out
the
Sun
suggestive
possible
and
understandings,
but
it's
very
hard
to
to
write
it
down
as
code
already
time
as
equation,
so
you
can
take
that
away
with
you.
B
It
doesn't
move,
it
doesn't
move
easily
or
chiefly,
it
cannot
be
moved
easily
and
cannot
be
scaled
up
easily.
So
we
have
to
have
this
webinar
have
this
video
and
in
a
better
situation
we
need
to
get
a
coffee
and
sit
down
and
talk
in
person
about
this
complicated
issue
right,
because
this
is
very
complicated.
B
It's
very
common
contact
space
in
the
in
the
other
end,
if
we
talk
about
higher
knowledge,
its
contact
when
available,
which
means
that
you
can
codify
that
you
can
crystallize
that
you
can
stabilize
it
and
you
can
write
on
as
textbooks
in
tech
schools
in
equations
in
coasts.
You
move
that
you
take
that
away
with
you
across
time
and
space,
so
you
don't
worries
you
don't
need
to
cite
the
Newton
anymore,
because
everything
you
need
to
know
about.
Newton
is
in
the
modern
text
code
of
physics,
but
use
you.
B
You
probably
still
need
to
cite
I,
don't
know
confused
little
bull
but
by
the
disciple
of
Confucius
again
and
again,
you'll
get
excited
by
border
again
again,
why
weather
is
there's
no
such
thing
about
like
text
or
the
Bible
in
the
standardized
Knology
in
such
a
way
that
you
don't
need
to
cite
referred
Bible
again,
because
the
knowledge
in
Bible
is
soft
norwich,
it's
not
like
gravity,
it's
very
context
based
and
you
haven't
always
go
to
the
context.
You
have
actually
have
to
be
review
the
context
to
really
understand
what
you
just
mean.
B
What
do
you
say?
So
these
are
the
two
alternative
measures
that
I
want
to
present
to
you.
One
one
measure
is
like
now.
We
know
the
sovereign
ology
cannot
is
expensive
to
supply
means
that
it
cannot
be
easily
skill
up.
It
takes
many
many
many
minutes,
many
many
hours,
many
many
days
to
communicate,
to
discuss
these
kind
of
expensive
issues
and
topics
to
for
to
have
a
student
like
leadership
to
have
children
like
know
how
to
negotiate
so
to
supply
this
institution.
It's
expensive.
B
You
need
to
hire
good
faculties
with
good,
with
with
rich
experience,
so
so
one
assumption
we
can
made
about
that
is
even
sub
knowledge
is
expensive
to
supply.
Then
you,
if
we
do
the
correlation
between
the
discipline
ranking
versus
the
college
rankings,
which,
on
x-axis,
which
is
highly
dependent
on
the
available
research
resources
like
money,
resource
space
resource
and
whether
you
are
close
on
an
ear
to
culture
and
intellectual
center
open
seriousness,
and
that
these
kind
of
things
will
determine
the
the
general
ranking
of
a
university
right.
You
have.
B
These
top
universities
have
rocks
for
all
these
universities,
so
the
X
is
now
really
kind
of
index
of
like
how
much
resource
you
have
and
how
much
expense
I
mean
it's
absolute
like
you
can
see
that
how
it
directly
relates
to
the
tuition
fee.
You
need
to
pay
to
get
in
these
kind
of
institutions.
Now
then,
like
you,
have
you
can
have
the
discipline
ranking
either
is
a
communication
it
is
sociology
at.
B
There
is
political
science,
always
engineering,
physics,
mathematics,
computer
science
and
then
here
what
I
did
is
I
analyze
their
correlations
and
separated
into
four
groups?
And
now
you
have
you,
you
can
cheat
directly
like
in
most
of
the
knowledge
domains.
The
knowledge
is
the
correlation
between
these
two
is
high.
You
have
a
higher
the
Pearson
correlation,
so
I
think
the
Spearman
correlation.
Is
you
see
it
like
in
the
illa
in
the
morning,
her
knowledge
in
in
mathematics
and
physics
and
computer
science
even
have
a
low
correlation?
B
That's
because
you,
this
kind
of
knowledge
are
easier
to
move
that
you
can
you
can.
You
can
build
a
good
mathematics
program
without
spending
with,
like
other
physics
or
engineering
program,
without
it
installing
a
lot
of
resources,
so
this
is
one
way
to
measure
it,
and
the
other
way
is
a
very
interesting
line
of
research
from
the.
B
Inevitability
of
science,
so
this
is
like
one
interesting
people:
I
want
to
pour
your
attention
to
is
Merton
stable
in
60s
about
multiples
and
Singleton's,
so
biggie.
The
question
asked
here
is
how
to
works.
Then
is
science
inevitable,
because
so
how
to
measure
that
one
way
to
measure
that
sizing
is
highly
available,
that
almost
every
scientific
discoveries
have
their
alternatives
like
if
I
stand
never
come
out
with
the
relative
theory,
then
there
will
be
other.
Other
people
come
up
with
similar
things
they
had
to
have
their
duplications.
They
have
their
alternative
copies.
B
So
if
we
lost
this
copy
as
a
human
society,
we
still
have
another
one.
So
there,
a
pioneering
work
done
by
Burton
in
1960s
and
inform
us
in
twenties
is
that
they
collect
the
whole
list
of
discoveries
and
calculate
the
probability
distribution
of
number
of
molecules
which
he
means
like
for
this
certain
invented
invention
or
discovery
how
much
copies
they
they.
B
They
have
only
how
much
people
claim
to
to
come
out
with
this
discovery
or
inversion
independent
without
coordinating,
which
other
and
very
interesting
thing
you
can
see.
That
is
like
a
lot
of
inventions
and
discoveries
are
multiples,
which
means
that
they
are
not
so
unique,
and
then
they
only
they
also
concentrate
in
time.
That
is
in
the
audio
track
degree
decades.
B
In
other
centuries,
most
of
molecules
of
scientific
discoveries
and
inventions
will
be
claimed
within
10
years,
like
70%,
without
within
ten
years
in
30%,
maybe
within
one
year,
and
people
also
find
that
this
is
accellerating.
Now
people
are
like
coordinating
synchronizing
and
if
you
don't
crane
your
scientific
discoveries
immediately,
then
a
couple
of
weeks,
maybe
it's
some
other
people
will
cremate.
That's
why
we
are
also
anxious,
as
as
a
scholar
and
a
very
interesting
thing
based
on
this
measure
is
like
is
the.
B
If
there's
a
dispute
like
if
we
multi-divisional
some
dispute
with
mean
meal
value
mean
value
of
me.
Meal,
that's
terrorized,
to
one
extent
on
average,
invention
or
thinking
or
idea
or
discovery
will
have
a
copy.
Then
this
value
is
it's
a
large
like
greater
than
1.
Then
it
means
the
always
have
a
like
alternative
copy
when
it's
smaller
than
1
means
it's
more
like
unique
and
you'll
find
that
hard
orange
Hardy's
friends
like
mathematics.
B
This
is
these
various
larger
and
then
for
biology,
which
is
like
more
contacts
based
fields,
it's
it's
smaller,
so,
which
means
like,
if
you
train
the
naive
ways
feels
like
an
origin
feels
like
mathematics
is
more
like
inevitable.
If
you
don't
come
rather
with
recent
proof,
other
people
will
come
out
with
this
fool,
but
then,
like
biology
is,
like
is
probably
less
predetermined.
We
less
inedible.
So
this
is
another
perspective
to
look
at
the
soft
Ohio
Ridge,
what
it
is
not
just
like
the
cost,
but
also
the
alternatives.
B
Ok,
so
sorry
for
a
long
narrative,
but
I
really
want
to
engage
you
in
this
like
very
interesting
content
that
I'm
dealing
with
and
then
I
can
share
share
with
you
a
little
bit
of
preliminary
findings
that
what
we
do
to
to
relate
to
associate
debating
and
novelty.
So
we
are
signing
from
the
we
are
building
our
work
from
a
great
paper
by
a
group
of
leading
leading
by
10
JavaScript
from
Stanford,
and
so
what
they
are
doing
lets
you
from
the
left
figures.
B
You
can
see
that
we
people
have
been
analyzing
citation
on
law
right.
We
always
assume
that
citation
is
some
kind
of
endorsement,
although
we
know
it,
but
our
heart
is
not
is
better
in
humanity
and
so
social
science.
Many
times
we
cite
people
to
debate,
to
disagree
right
so
in
in
their
in
their
pioneering
research.
They
try
to
have
a
machinery
model
to
identify
the
function
of
citation.
As
you
can
see
from
this,
this
bars
a
citation
and
also
from
three
examples
from
the
the
contest
they
provided.
B
You
can
cite
something
somewhere
something
or
someone
to
contrast
to
compare.
Well,
you
can
size
someone
to
use
to
build
your
research
account
or
you
can
assign
someone
as
a
as
a
shooting
lights
on
the
landscape
environment,
so
society
as
a
background.
Also,
you
can
inside
too,
as
to
show
that
your
research
is
too
light
as
a
motivation.
B
So
there's
extensions
as
future
work.
That
kind
of
thing,
so
you
have
different
kind
of
functions
from
the
citation.
If
you
look
at
into
the
context,
but
not
just
common
members,
not
just
common
the
length
of
this
different
least,
your
reading
abilities
can
specify
it's
why
people
are
citing
this
literature
or
scientists
call
of
four
and
the
sunday
purpose
on
this
machine
learning
model.
They
can
identify
the
functions
and
they
have
some
interesting
perseverance
like
from
formal
publications,
like
journals
all
the
way
to
informal
publication.
B
Like
workshops,
people
are
talking
what
about
what
we
are
thinking
as
a
convention.
It,
like
you,
cite
as
a
background
to
what
I
am
doing
as
author
to
side
as
uses,
and
also
they
analyze
a
main
conference
ACR
in
the
computational
linguistics
in
the
past
decades,
and
they
found
that
more
and
more
people
are
citing
for
use.
Less
and
less
people
are
citing
4
compare
and
construct.
So
he
is
in
the
field.
The
label
observing
science
is
getting
more
pratical
and
less
debatable
people.
People
don't
want
to
debate.
B
People
don't
want
to
compare
people
don't
want
to
challenge
each
other.
People
just
want
to
have
one
more
piece
to
extend
a
little
bit
to
build
a
little
bit
to
use
it
a
little
bit.
They
are
more
like
more
like
engineering
style,
rather
than
like
refractive
style,
keep
going
on,
at
least
in
computer
science
in
one
domain
durable
observing
and
we
are
using
them,
then
machine
learning
model
for
actually
we
did
a
little
bit
of
revision,
including
the
neural
network
phase,
and
we
included
a
new
model
called
transformer
in
terms
of
to
improve
the
performance.
B
But
I
will
leave
these
technical
details
on
the
back
and
well
after
we
improve
that
citation
function,
identifying
machine
learning
model
we
use
a
to
analyze.
Whether
debate
in
science
also
contribute
to
novelty.
There's
a
one
dimension
of
knowledge.
We
define
or
care
that
that
is
so.
Basically,
we
can
select
the
set
of
article
and
then
we
can
separate
them
into
based
on
their
because
in
computer
science,
the
fabrication
manual
as
a
conference
is
the
most
important.
B
Sorry
I
think
I
think
I
made
a
wrong
that
it's
yeah
annotations
right,
so
a
CRS,
more
context
available,
because
it's
more
about
the
general
rules
and
algorithms
it's
more
like
harmonic
and
then,
like
the
the
other,
like
II,
MMP
NLP,
the
e/m
is
merely
for
empirical
methods.
So
it's
for
more,
like
context
relevant
more
tomorrow,
massively
more
apply
consequence
of
linguistic
computational
linguistic
models
rather
than
the
theoretical
discussions.
B
B
Yes,
we
observe
the
universal
correlation
between
towards
then
this
article
is
using
citations,
it's
exciting
for
compare
challenge
and
debates
as
an
axe,
as
you
can
see
from
our
axes
and
the
performance
verbal
that
the
order
depend
on
forego
in
a
way
that
to
extent
the
important
keywords
identified
by
tf-idf,
which
is
like
such
ng
index
to
to
locate
in
unique.
We
presented,
which
resonated
keywords
for
abject
the
works.
Then
this
kind
of
keywords
were
absent
and
more
nobly
combined
dates
rather
than
conventionally
combined
it.
So
it's
like
it's
a
measure
of.
B
Is
this
article
talking
about
something
new,
some
new
associations
between
important
things
or
some
like
some
combination
of
knowledge
that
remains
lovely
and
under
export?
So
the
answer
is
yes,
the
answer
is
yes,
you
already
see
that
this
kind
of
positive
correlation-
if
you
tell
any
more,
if
you
debate
more,
then
you
are
creating
assumptions
or
in
findings
more
hidden
and
the
discover,
links
between
knowledge,
so
you're
contributing
to
knowledge,
novelty
and
also
very
interesting
me.
B
We
do
also
observe
that,
in
soft
more
like
contacts
based
knowledge,
you
need
to
be
paid
more
to
create
novelty
compared
to
the
more
contacts
irrelevant
Hamrick
using
these
using
these
two
conference.
As
an
example
for
the
for
Imperial
method
of
neural
linguistic
models,
then
people
are
more
likely
to
debate
and
then,
like
more
negative.
Things
are
creates
in
the
modern
novel
knowledge
you
are
created
in
there
in
there
and
compared
to
male
contact
developments.
B
The
values
of
application
people
are
less
likely
to
debate,
and
then,
let's
nobody's
credits,
I
mean
even
you
think
of
the
the
percent
house
in
terms
of
like
the
rankings,
then,
with
the
same
amount
of
debates,
people
looks
like
to
create
more
novelty,
more
surprise,
more
unknown
knowledge
in
soft
knowledge,
rather
than
car
knowledge.
So
this
is
like
we
address
these
two
questions
in
the
in
context.
B
We
just
introduced
and
then
like
provide,
is
like
very
pretty
very
but
very
interesting
results
and
I
will
leave
most
of
the
time
for
the
for
the
discussion
so
that
we
have
we
can
unfold.
This
very
interesting
topic
fully,
but
I
like
to
take
this
opportunity
to
acknowledge,
like
all
great
collaborators
who
also
contribute
to
the
business
and
the
sound
of
results
and
also
have
exciting
projects
going
on
the
first.
B
And
we
are
looking
at
the
millions
of
law
cases
and
try
to
see
how
this
kind
of
debate,
how
this
kind
of
debate
not
about
the
knowledge,
but
also
about
the
values,
change
and
structured
our
society
and
how.
For
example,
the
attendant
is
kewal
as
used
to
be
a
crime.
And
now
it's
not
a
cry
anymore.
And
when
did
this
happen
and
how
this
is
this
happen,
together
with
other
values,
with
anchoring
values
in
our
collectively
cognitive
system
as
a
law
and
how
the
institutional
things
and
how
these
things
may
change.
B
So
that
is
the
great
interesting,
exciting
project
going
on
with
an
angel
from
University
of
Chicago,
the
common
sociology
and
then
the
last
thing
about
a
debate
we
are
doing
in
science.
They
were
looking
at
the
whole
science
fear
evolution
over
more
than
100
years
with
available
data.
We
look
at
the
40
43
million
articles
and
one
of
10
million
scholars
and
see
how
this
debate
happens.
B
B
Maybe
it's
not
always
get
stuck
because,
like
it's
still
too
far
too
early
to
jump
to
the
conclusion
whether
this
is
totally
a
bad
thing
or
it
could
be
also
a
good
thing,
but
you
see
how
people
build
up
their
identity
when
they
became
independent
after
the
adolescence
middle
jump
tool
to
tighten
more
recent
reference,
then
already
happily
kind
of
Judah,
but
also
get
this.
This
identity,
this
culture
and
Nora's
identity
get
sticky
and
then
keep
living
in
the
world.
B
C
C
Sometimes
you
have
articles
that
bring
conspiracy
theories
or
there
is
a
hard
to
reach
consensus
on
where
it
is
climate
debate
or
whether
it
is
vaccination
versus
autism
debate.
So
has
your
work
or
someone
else's
work
shed
light
on
those
type
of
articles
where
there
is
a
lot
of
debate
but
hard
to
imagine
that
they
will
be
new
knowledge
created
yeah.
B
That's
that's
a
great
great
question,
that's
also
the
question.
I
could
be
thinking,
and
so
there
are
many
points
and
I
I
didn't
the
first
one
I
I
want
to
I'm
totally
agree
all
right.
You
see
people,
you
see,
people
with
which
debate
like
so
you
see
the
debates,
looks
at
I,
never
going
anywhere
right
and
you
see
people
who
never
convinced
by
evidence
effect
from
the
other
sides.
Actually,
one
quick
one
phenomenon
we
open,
always
not
easy
new
social
media
is
that
people
using
conflicting
evidence
as
important
for
the
whole
steamer
right.
B
So
that's
that's.
There's
also
like
one
more
abrasion
behind
the
aging
dynamics
of
what
I'm
doing
like
it's
a
lot
of
times
is
rather
angels
like
the
debate
phonology,
but
people
like
debate
for
identity,
but
then,
when
it
comes
to
identity
issues,
it's
very
hard
to
to
completely
change
people
right,
because
what
they
are
talking
about
is
really
about
or
what
I
think
the
word
is
look
like,
and
so
that,
but
but
we've
versed
with
the
technology
of
that
kind
of
debate
and
going
on
with,
which
is
how
to
converge
rather
than
progressive.
B
This
is
like
a
dominant
principle
that
you
cannot
save
memory
and
tiny
both
at
the
same
time.
For
example,
if
you
want
to
print
out
1
million
digital
pie,
then
you
have
two
ways:
you
can
minimize
your
memory.
You
write
a
small
program
that
then,
that
program
will
continue
as
it
compute
when
they
print
it
out.
In
that
way,
you
say
it's
safe
memory
and
space,
but
you
cost
time
and
the
other
way
is
like
you,
pre
compute,
all
these
100
million
1
million
a
digital
file
and
put
it
somewhere
in
your
memory.
B
Now,
whenever
you
want
to
have
the
100
digits,
the
10,000
digits,
you
can
email
it
prenatal
out
in
this
way,
you
minimize
time,
but
you
consume
a
lot
of
memory
that
so
I
to
exact
thing.
I
would
suggest
our
human
or
social
brains,
its
simplicity.
You
cannot
minimize
time
complexity
and
memory
capacity
in
the
same
time
and
this
revisits
our
small
changes.
B
What
paper
published
on
nature
this
year,
that's
with
a
small
team
and
making
it
more
like
bigger
jaws
in
terms
of
cognitive
level
like
they
describe
existing
thinking
the
theories
and
findings
when
large
teams
were
preferred
to
endorse
it,
support
it
available
by
that
from
there
and
another
conclusion.
Another
thing
we
can
look
around
here
is
like
small
things
am
I
paying
more
time.
Complexity
lately
have
they
and
they
have
less
further
in
time
because
of
they
consume.
There's
money,
and
we
all
know
when
you
have
a
lot
of
money.
B
When
you
apply
a
lot
of
grants,
we
award
a
lot
of
grants
in
science
how
much
what
kind
of
pace
of
life
you
living,
but
you
have
to
keep
keeping
our
people
keep
delivering
things.
So,
in
that
sense,
going
back
to
what
we
asked
you
or
we
were
talking
about
is
like
sign,
is
used
to
make
Li
shuo
more
time
complexity,
and
for
that
issues
it
probably
would
take
a
very,
very
long
time
to
converge
and
to
find
answers.
B
A
B
Do
I
do
think
debate
will
push
for
knowledge
and
understanding,
but
the
president
debates
is
is
a
complicated
thing
in
a
way
that
it's
it's
a
it's
a
media,
behavior
right,
it's
it's
on,
it's
our
media
and
so
we
Kiwi,
so
there
are
different
audience
over
there
and
a
debate
may
not
only
push
forward
the
knowledge
from
one
group.
So
so
this
is
like
a
very
interesting,
but
they
realize
more
time
compressed.
Be
talking
then
anyway,
like
what's
the
consequence
and
what's
a
function.
B
What
should
be
the
go
of
this
political
debates,
but
one
one
consequence
I
can
see
from
there
is
like
political
debates,
push
for
knowledge,
two
different
social
groups
and
make
them
acknowledge
the
existence
of
each
other
right.
So
we
I
I,
see
a
lot
of
concern
and
a
shame
and
a
lot
of
these
concerns
like
there's
a
evil,
ongoing
polarization
and
then
divided
going
on.
But
then
I
was
again
going
to
the
opportunistic
point
of
view.
I
was
thinking.
Maybe
we
are
not
more
divided
than
before.
B
This
kind
of
debate
may
contribute
to
knowledge
in
terms
of
they
converge
ideas,
but
at
least
it
I
do
see
its
contribution
in
terms
of
unfolding
ideas,
because
the
science
is
at
least
for
the
follow
for
the
fields
that
I'm
more
comfortable
and
welcome
with
about
that
is
the
science
science
is
not
only
about
convergence,
it's
also
about
140.
Whenever
the
time
comes,
the
paradigm
shift.
It's
always
like.
B
We
people
try
to
compress
things:
people
have
a
lot
of
opposition
and
the
data
they
want
to
write
our
equations,
that
the
shorter
is
the
equation
as
possible
to
summarize
everything
they
know
and
then
equation
and
their
paradigm,
and
that
world
feel
will
last
for
decades
or
even
centuries,
until
people
find
conflicting
evidence,
people
and
then,
of
course,
it's
always
the
identity
thing.
We
are
all
human
beings.
So
there
comes
with
the
completing
evidence.
B
There
will
be
new
generations
and
the
completing
in
social
groups
and
then,
like
the
crucible
debates,
then
they
open
new
paradigms
that
with
more
more
powerful
in
summarizing,
more
complicated
data
in
the
shorter
equations
and
but
but
but
unfolding
and
challenging
and
making
a
breaking
down
the
existing
uniform
of
you
is
probably
the
cornerstone
of
the
next
step.
That's
just
some
quick
thinking,
but.
A
B
So
there
are
like
multiple
ways
and
the
trip,
but
dirty
way
is
you
select?
For
example,
if
you
have
this,
this
world
investigates
and
then
you
we
identified
the
I'm
speaking
for
the
others,
I'm,
not
otherwise
Liberto,
so
we
identify
its
alternatives.
For
example,
study
exam
analyses.
So
these
are
three
alternative
of
the
word
investigate.
Now
we
will
select
all
a
sentence
with
this
world.
This
eight
one
of
these
four
words,
as
contains
these
words
right
and
then
like.
We
will
identify
the
sentence
busy
like
in
a
very
strict
standard.
B
We
can
just
select
the
order
sentence
with
only
this
world
change.
For
example,
gender
is
like
we
investigate
data.
Now
we
select
these
sentence
now.
Whenever
next
time
you
observe
the
sentence
like
we
study
the
data
or
we
examine
the
data.
Well,
we
analyze
the
data.
Then
we
select
all
these
sentence.
I
will
put
in
together.
For
example,
there
are
100
cases
and
then
30
of
these
100
cases
is
using
investigate
and
then
90
is
using
Sally
6-4
sorry.
This
does
not
wanna
get
but
basic.
B
That's
exactly
the
example
that
you
have
different
frequencies
attached
to
different
words.
Now
you
can
calculate
the
entropy
or
the
variance
of
variance
anything
you
like
to
quantify
the
concentration
of
weight
in
this
particular
world,
for
example
in
mechanical
we
were
with,
although
we
cannot
even
identify
any
context.
This
use
exchange
burry
with
other
words,
but
for
the
for
the
investigate
example,
you
identify
a
lot
of
same
contents,
the
same
context
they
used,
but
with
a
replaced
word.
So
that's
a
way
to
to
to
measure
in
in
the
in
the
knife
battery
me.
B
There
are
also
other
ways
that
I
am
developing
using
neural
network
embeddings,
in
a
way
that
we
can
use
this
neuron
able
to
identify
language
complements
or
substitutes,
for
example,
Oxford,
University
and
Cambridge
University.
There
are
substitutes,
because
people
use
people,
people
talk
about
universities
and
universities
to
go
to
in
England,
so
they
like
the
kind
of
occupied
as
similar
position
in
the
context,
but
they're
like
Cambridge,
University
and
expensive
tuition
fee.
These
are
not
substitute.
B
These
are
compromised
because
they
are
did
they
are
similar
with
each
other
and
that
you
have
to
put
them
together
to
to
phone
a
meaning
to
make
it
a
function.
So
there
are
what
are
embedding
technique
to
identify
it.
This
kind
of
complements
and
substitutes
of
words
in
a
more
Prada,
more
set
of
contacts,
but
what
they
are
doing
as
author
in
this
paper
is
really
just
at
any
viral
context,
calculate
the
weights
and
it
calculates
the
concentration
I.
B
In
terms
of
so
the
way
they
develop
it
in
these
people
is
how
a
view
is
even
more
handcraft
and
commutation
in
terms
of
algorithms.
It's
community,
it's
easier.
It's
simple,
it
just
I
think
if
I
were
the
words
or
the
sentence
and
they
calculate
the
words,
but
it's
very
handcraft.
You
have
to
identify
the
words
you
want
and
identify
other
alternatives
and
then
identify
order
of,
but
I
would
probably
suggest
that,
wouldn't
very
it's
actually
a
compared
to
this
matter.
B
So
that's
a
very
tender
on
a
very
huge
amount
of
data
and
they
provide
a
long
list
of
1
billion
words.
Each
word
is
representing
300
dimensions
and
by
analyzing
that
data,
let's
share
the
parameters,
is
also
a
cheap
way
to
to
identify
it's
world
compliments
and
substitutes.
So
it's
like
the
famous
woman
miners,
main
press
key
equals
Queen
example.
You
can
identify
these
hidden
dimensions
by
gender
and
power
and
then
like
see
how
replaceable
or
using
interchangeable
tours
are.