Adam Wyner's presentation on natural language processing of argumentation such as found in social media, newspapers, and law. Relevant to semantic web, text analysis, computational linguistics, argumentation. University of Aberdeen.
Natural Language Processing of Argumentation. Adam Wyner. Summer School Lecture 2014
1. Argument
Extrac.on
from
Social
Media
Using
GATE
Adam
Wyner
Compu.ng
Science,
University
of
Aberdeen
Summer
School
on
Argumenta.on:
Computa.onal
and
Linguis.c
Perspec.ves
University
of
Dundee
Sept.
7,
2014
2. Goals
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
• Iden.fy
materials
(social
media)
and
generic
issues.
• Outline
linguis.c
issues.
• Outline
GATE
methodology.
• Provide
some
examples.
2
4. Where
Arguments
Appear
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
4
• Consumer
websites:
Amazon,
eBay,...
• Law:
policy
making,
Supreme
Court
transcripts,
case
based
reasoning,
regula.ons.
• BBC's
Have
Your
Say
and
Moral
Maze.
• Medical
diagnosis.
• Current
events.
• Making
plans.
• Debatepedia,
Wikipedia,
mee.ng
annota.ons,
web-‐forums,...
• Social
media:
Facebook,
da.ng
5. Current
Events
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
• ScoZsh
Independence
and
Currency
• h[p://www.bbc.co.uk/news/uk-‐scotland-‐scotland-‐
poli.cs-‐2622589
5
6. ScoZsh
Independence
2014
The
issue
of
what
currency
an
independent
Scotland
would
use
has
become
the
key
ba[leground
of
the
referendum
debate.
The
ScoZsh
government
is
in
favour
of
a
sterling
zone,
saying
it
would
be
in
the
interests
of
both
Scotland
and
the
UK
to
con.nue
to
formally
share
the
pound
if
the
former
votes
for
independence,
ensuring
stability
for
both
states.
UK
chancellor
George
Osborne
has
said
the
UK
would
not
enter
into
a
currency
union
with
Scotland
if
it
voted
'Yes'
in
September's
referendum,
claiming
such
a
union
would
be
against
the
economic
interests
of
England,
Wales
and
Northern
Ireland.
Mr
Osborne's
statement
was
the
UK
government's
strongest
interven.on
in
the
debate
yet,
and
his
posi.on
was
supported
by
both
Labour
and
the
Liberal
Democrats.
First
Minister
Alex
Salmond
countered
Mr
Osborne's
claims
in
a
speech
to
pro-‐independence
business
leaders
in
Aberdeen
on
Monday,
which
he
said
had
"deconstructed"
the
case
against
a
currency
union.
So
what
are
Mr
Osborne's
key
arguments
and
how
has
Mr
Salmond
sought
to
counter
them?
Claim:
Trade
with
Scotland
is
important
to
the
UK,
but
the
overall
propor;on
is
small
George
Osborne:
"I'm
the
first
to
say
that
our
deeply
integrated
businesses
and
their
suppliers
are
compelling
reasons
for
keeping
the
UK
together
-‐
70%
of
ScoZsh
trade
is
with
the
rest
of
the
UK.
That
is
a
massive
propor.on.
"And
trade
with
Scotland
is
important
to
the
rest
of
the
UK
-‐
but
at
only
10%
of
the
total
trade,
it
is
a
much
smaller
propor.on.
These
trade
figures
don't
make
the
unanswerable
case
for
a
shared
currency
that
the
ScoZsh
government
assume."
Alex
Salmond:
"I
am
publishing
an
es.mate
of
the
transac.ons
costs
he
would
poten.ally
impose
on
businesses
in
the
rest
of
the
UK.
They
run
to
many
hundreds
of
millions
of
pounds.
My
submission
is
that
this
charge
-‐
let
us
call
it
the
George
tax
-‐
would
be
impossible
to
sell
to
English
business.
"In
fact
if
you
remove
oil
and
gas
from
the
equa.on,
Scotland
is
one
of
the
very
few
countries
in
the
world
with
which
England
has
a
balance
of
trade
surplus."
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
6
7. Arguments
in
debategraph.org
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
Current
Method
-‐
read
text
-‐
manually
analyse
-‐
manually
enter
text
into
tool
-‐
manually
annotate.
Problems
-‐
slow,
costly,
error-‐
prone,
ad
hoc,
must
search
for
'place'
of
new
addi.ons,
etc....
7
8. Consumer
Comments
on
Amazon
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
8
9. Pro
and
Con
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
9
10. Comments
on
Comments
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
10
11. Policy
Consulta.ons
-‐
LIBER
on
Copyright
-‐
Ques;on
9.
Should
the
law
be
clarified
with
respect
to
whether
the
scanning
of
works
held
in
libraries
for
the
purpose
of
making
their
content
searchable
on
the
Internet
goes
beyond
the
scope
of
current
excep;ons
to
copyright?
-‐
Yes.
-‐
Not
all
the
material
digi.sed
by
publishers
is
scanned
with
OCR
(Op.cal
Character
Recogni.on)
with
the
purpose
of
making
the
resul.ng
content
searchable.
If
the
rights
holders
will
not
do
this,
libraries
should
be
able
to
offer
this
service.
It
would
have
a
transforma.ve
effect
on
research,
learning
and
teaching
by
opening
up
a
mass
of
content
to
users
which
can
be
searched
using
search
engines.
The
interests
of
copyright
holders
will
not
be
harmed,
because
the
resul.ng
output
will
act
as
marke.ng
material
for
their
materials.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
11
12. What
Needs
to
be
Done?
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
• Annotate
textual
passages
for
argument
relevant
por.ons
(premise,
claim)
• Annotate
rela.ons
amongst
passages
(premise
of
what
argument)
• Represent
in
some
machine
readable
form.
• Thought
experiments
to
objec7fy
and
abstract
the
issues.
12
13. Generically
What
Needs
to
be
Done?
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
13
14. What
Needs
to
be
Done?
Basic
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
14
andalka
nadlka
fa
adlkaa
la
lkd
alkdj
a
a
akj
dal;k
fda
ada.
a;lkd
a
andalkda
anda;k
a
jad
ie
ae.
a;lkd.
nainea
;
alkei
nai
lalin
oa
nekn.
ake
anoiiena
dk
aieane0-‐a
an
a;kl
aeu
ajena.
;oi
anoi
alkd
ao;na
oen
oiana
oin.
l
;kja
dka
j
ajda
djflka
kle
ak
kad
la
ien
ae
n
en.
lkj
ad
ad
fa
;adja
dfakd.
Source
Text
A
a1.p.1.
-‐
andalka
nadlka
fa
adlkaa
la
lkd
alkdj
a
a
akj
dal;k
fda
ada.
a1.p.2.
-‐
a;lkd
a
andalkda
anda;k
a
jad
ie
ae.
a;lkd.
a1.c.
-‐
nainea
;
alkei
nai
lalin
oa
nekn.
a2.p.1
-‐
ake
anoiiena
dk
aieane0-‐a
an
a;kl
aeu
ajena.
a2.p.1
-‐
;oi
anoi
alkd
ao;na
oen
oiana
oin.
a2.e.3
-‐
l
;kja
dka
j
ajda
djflka
kle
ak
kad
la
ien
ae
n
en.
a1.c
-‐
lkj
ad
ad
fa
;adja
dfakd.
Annotated
Text
A
Key:
premise,
excep.on,
claim
15. What
Needs
to
be
Done?
Ques.ons
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
• How
do
we
know
(as
readers)
which
is
a
premise,
which
is
a
claim,
and
which
is
an
excep.on?
– explicit
linguis.c
markers
(e.g.
assuming
X,
therefore
Y)
– order
of
sentences?
– other,
e.g.
context?
• If
we
scrambled
the
order
of
the
sentences,
could
we
recons.tute
the
argument
annota.on?
– Engineer
–
"Doesn't
happen,
not
relevant.
Build
for
par.culars."
– Scien.st
–
"Does
it
happen?
If
it
does
or
could,
how
do
we
address
it?
Explore
for
principles."
15
16. What
Needs
to
be
Done?
Scramble
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
16
andalka
nadlka
fa
adlkaa
la
lkd
alkdj
a
a
akj
dal;k
fda
ada.
nainea
;
alkei
nai
lalin
oa
nekn.
a;lkd
a
andalkda
anda;k
a
jad
ie
ae.
a;lkd.
;oi
anoi
alkd
ao;na
oen
oiana
oin.
lkj
ad
ad
fa
;adja
dfakd.
l
;kja
dka
j
ajda
djflka
kle
ak
kad
la
ien
ae
n
en.
ake
anoiiena
dk
aieane0-‐a
an
a;kl
aeu
ajena.
Source
Text
A
a1.p.1.
-‐
andalka
nadlka
fa
adlkaa
la
lkd
alkdj
a
a
akj
dal;k
fda
ada.
a1.p.2.
-‐
a;lkd
a
andalkda
anda;k
a
jad
ie
ae.
a;lkd.
a1.c.
-‐
nainea
;
alkei
nai
lalin
oa
nekn.
a2.p.1
-‐
ake
anoiiena
dk
aieane0-‐a
an
a;kl
aeu
ajena.
a2.p.1
-‐
;oi
anoi
alkd
ao;na
oen
oiana
oin.
a2.e.3
-‐
l
;kja
dka
j
ajda
djflka
kle
ak
kad
la
ien
ae
n
en.
a1.c
-‐
lkj
ad
ad
fa
;adja
dfakd.
Annotated
Text
A
Key:
premise,
excep.on,
claim
17. Scramble
in
Comment
Update
Argumenta.on
Summer
School,
Dundee
nada
dnana
a
kkkd
andai
;a.
n=jja
nmae
a;kda
nIanl.
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
17
andalka
nadlka
fa
adlkaa
la
lkd
alkdj
a
a
akj
dal;k
fda
ada.
a;lkd
a
andalkda
anda;k
a
jad
ie
ae.
a;lkd.
nainea
;
alkei
nai
lalin
oa
nekn.
Source
Text
A
a1.p.1.
-‐
andalka
nadlka
fa
adlkaa
la
lkd
alkdj
a
a
akj
dal;k
fda
ada.
a1.p.2.
-‐
a;lkd
a
andalkda
anda;k
a
jad
ie
ae.
a;lkd.
a1.p.3.
-‐
n=jja
nmae
a;kda
nIanl.
a1.e.2.
-‐
nada
dnana
a
kkkd
andai
;a.
a1.c.
-‐
nainea
;
alkei
nai
lalin
oa
nekn.
Annotated
Text
A
plus
Key:
premise,
excep.on,
claim
Source
Text
B
Source
Text
C
a;lkd
a
andalkda
likalaka
anda;k
a
jad
ie
ae.
a;lkd.
(contrary
to
a1.p.2)
Source
Text
D
18. Scrambling
Ques.ons
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
• How
do
we
know
when
two
premises
are/are
not
the
same?
• How
do
we
know
what
argument
to
a[ach
a
proposi.on
to?
• Addressing
these
ques.ons
may
require
some
deep
syntac.c
and
seman.c
analysis
(hint,
I
think
it
does
and
can
be
done....eventually).
• BUT
VERY
HARD!!
• Find
a
less
demanding,
near
term
approach
towards
similar
objec.ves.
18
19. Generic
Issues
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
• Reconstruc.on
of
arguments
from
textual
sources:
– extrac.on
for
argument
evalua.on.
– Discon.nuity
of
arguments
in
textual
source.
– Knowledge
base
construc.on
and
dynamics.
• Linguis.c
issues:
– Domain
terminology.
– Linguis.c
informa.on
and
variety
(many-‐to-‐one
sentence-‐
proposi.on).
– Argument
rela.ons
(premise,
claim,
excep.on,
contrary).
– Sources
of
defeasibility
(epistemic
'strength').
– Other
argument
component,
e.g.
proposi.onal
aZtudes
(e.g
believe,
know),
speech
act
verbs
(e.g.
assert,
grant).
19
21. Loca.ng
the
Problem
and
Engineering
a
Solu.on
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
21
• The
knowledge
acquisi.on
bo[leneck
from
NL
to
some
formal
representa.on.
• Rela.onship
to
other
parts
of
the
argumenta7on
processing
pipeline.
22. Three
Stages
Graph
–
Structured
or
Instan.ated
AFs
gOkZI[jjQ][
gZIq]gX
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
22
[]qYIGOI
23. hI rjI[hQ][h]N
gOkZI[jh
rjI[hQ][h]N
][EYkhQ][h
/jIdÂE][hjgkEj
gOkZI[jh[GjjEXh
/jIdÃQGI[jQNshIjh]N
EEIdjIGgOkZI[jh
/jIdÄQGI[jQNshIjh]N
EEIdjIGE][EYkhQ][h
Three
Stages
-‐
Caminada
and
Wu
2011
Knowledge
Acquisi.on
Bo[leneck:
.me,
labour,
exper.se
to
construct
a
KB
at
scale.
24. Logic-‐based
Instan.ated
Argumenta.on
Besnard
and
Hunter
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
23
• An
argument
is
an
ordered
pair
ψ,
α;
ψ
is
a
subset
of
a
given
KB
and
α
is
an
atomic
proposi.on
from
the
KB;
ψ
is
a
minimal
set
of
formulae
such
that
ψ
implies
α,
and
ψ
does
not
imply
a
contradic.on.
ψ
is
said
to
support
the
claim
α.
• Where
p
and
q
are
atoms,
and
where
the
KB
is
comprised
of
p
and
p→q,
then
{p,
p→q},
q
is
an
argument.
• We
could
have
a
KB
from
which
we
can
form
an
argument
which
supports
¬q,
{p,
p→¬q},
¬q.
In
addi.on
and
with
respect
to
this
argument,
suppose
we
can
form
an
undercuer
{r,
r→¬p},
¬p
and
a
rebual
{r,
r→¬p,
¬p→q},
q}.
• KBs
(even
rela.vely
small
ones)
generate
lots
of
arguments
and
a[ack
rela.onships
which
can
be
structured
in
a
tree.
25. Abstract
Argumenta.on
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
24
Preferred
extension:
{a,
c,
d,
h,
i,
k}
26. Zeroing
In
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
Source
text
Knowledge
base
argumenta.on
schemes
Generated
arguments
(abstract
or
instan.ated).
25
27. Context
with
Respect
to
Analysis
and
Argumenta.on
Schemes
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
26
28. Current
Tools
to
Extract
and
Structure
Arguments
from
Text
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
27
• Ra.onale,
Araucaria,
Carneades
(Gordon
2007),
IMPACT
Project,
Legal
Appren.ce,
Argument
Wall,....
• Pale[e
of
annota.ons
and
templates.
• All
manual.
No
NLP.
29. Argumenta.on
Schemes
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
• Pa[erns
of
presump.ve
(defeasible)
reasoning
(Walton
1996)
• Prac.cal
Reasoning
with
values:
– Do
ac.on
(transi.on)
because:
• Current
circumstances
-‐
a
list
of
literals.
• Consequences
–
a
list
of
literals.
• Values
(promoted,
demoted,
neutral
wrt
ac.ons)
–
a
list
of
terms.
• Credible
Source:
– Z
is
accepted
because:
• X
is
an
expert
in
domain
Y.
• X
stated
literal
Z
• Z
is
about
domain
Y.
28
30. Overall
Proposal
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
• Normalise
natural
language
source
material
into
argumenta.on
schemes.
• Formalise
argumenta.on
schemes
in
terms
of
roles
of
proposi.ons
in
the
scheme
and
internal
structure
of
proposi.ons
(predicates
and
typed
variables).
• Connect
argumenta.on
schemes
to
abstract
arguments.
• Relate
one
scheme
to
another
in
terms
of
contrariness.
• Extract
scheme
relevant
informa.on
from
the
source.
• Create
a
knowledge
base
to
instan.ate
variables.
29
31. Caveat
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
• Low
level
automa.on,
using
high
level
structures
as
guides.
• For
example,
no
automa.c
search
for
scheme
filling,
grounding
of
variables,
contrast
iden.fica.on.
• Progress
can
be
made
on
these
(and
for
contrast
iden.fica.on,
there
is
significant
work
already).
30
32. Normalise
for
Argumenta.on
Schemes
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
31
33. Annotate
–
Query
–
Extract
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
32
• Annotate
with
respect
to
Argumenta.on
Schemes.
– characteris.c
terminology
of
the
scheme.
– generalise
the
terminology
to
cover
varia.on.
– dis.nguish
domain
from
generic
terminology.
• Complex,
flexible
queries
over
the
annota.ons.
– Low
level
(atomic)
and
high
level
(molecular)
construc.ons.
– Interac.ve,
semi-‐automa.c.
• Export
to
some
machine
readable
format
-‐
XML.
35. Problems
with
Language
I
• Iden.fica.on,
implicit
informa.on,
mul.ple
forms
with
the
same
meaning,
the
same
form
with
mul.ple
meanings:
• En.ty
ID:
Jane
Smith,
for
plain.ff.
• Rela.on
ID:
Edgar
Wilson
disclosed
the
formula
to
Mary
Hays.
• Bill
drove
the
car
into
Phil
at
60
MPH.
(agent,
instrument,
killing)
• Jane
Smith,
Jane
R.
Smith,
Smith,
A[orney
Smith....
• Jane
Smith
in
one
case
decision
need
not
be
the
same
Jane
Smith
in
another
case
decision.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
34
36. Problems
with
Language
II
• Concepts,
dispersed
meanings,
rules,
diathesis:
• Plain.ff,
judge,
a[orney.
• Jane
Smith
represented
Jones
Inc.
She
is
a
partner
at
Dewey,
Chetum,
and
Howe.
To
contact
her,
write
to
j.smith@dch.com.
• If
a
woman
is
over
62
years
old
and
lives
in
the
UK,
she
is
a
pensioner.
• Diathesis:
alterna.ve
sentence
forms
with
(almost)
synonymous
meaning:
Bill
pushed
Jill;
Jill
was
pushed
by
Bill.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
35
37. Problems
with
Language
III
• Ambiguity,
vagueness,
underspecifica.on:
• The
man
saw
the
woman
with
binoculars.
• It
is
illegal
to
leave
a
heap
of
shoes
on
the
sidewalk.
• Vehicles
may
not
be
driven
in
the
park.
• Sarcasm,
irony.
• Interpreta.on.
• Context
dependence,
subjec.vity,
arbitrary
meaning,
when
I
was
at
school,
I
know
language....
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
36
38. Problems
with
Language
IV
• Complexity,
length,
and
layout
(see
our
Camera
example).
• Intersenten.al
connec.ons:
• Bill
le
the
house.
He
drove
home.
• Bill
le
the
house.
He
didn't
feel
comfortable
there.
• Bill
le
the
house.
It
was
an
old
house,
once
owned
by
a
wealthy
merchant.
• Synonymy,
antonyms,
meronyms
(finger
part
of
hand),
etc.
• Repe..on.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
37
39. Problems
for
Annota.on
• Annotate
large
legacy
corpora.
• Address
growth
of
corpora.
• Reduce
number
of
human
annotators
and
tedious
work.
• Make
annota.on
systema.c,
automa.c,
and
consistent.
• Annotate
fine-‐grained
informa.on:
• Names,
loca.ons,
addresses,
web
links,
organisa.ons,
ac.ons,
argument
structures,
rela.ons
between
en..es.
• Map
from
well-‐draed
documents
in
NL
to
RDF/OWL/XML.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
38
40. Addressing
the
Problems
• Decompose
big
problems
down
to
smaller
problems.
• Modularise
problems.
• Address
the
smaller,
modular
problems.
• Compose
solu.ons
from
parts.
• Iden.fy
(set
aside,
address,
assign
to
someone
else)
remaining
and/or
highly
problema.c
issues.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
39
42. Approaches
• Knowledge
light
in
terms
of
knowledge
of
the
domain
or
of
language
–
sta.s.cal
or
machine
learning
approaches.
Algorithmically
compare
and
contrast
large
bodies
of
textual
data,
iden.fying
regulari.es
and
similari.es.
Sparse
data
problem.
Need
a
‘gold
standard’.
No
rules
extracted.
Opaque.
Hard
to
modify.
• Knowledge
heavy
in
terms
of
lists,
rules,
and
processes.
Labour
and
knowledge
intensive.
Creates
gold
standards.
Transparent.
Can
jus.fy
outcomes.
Can
'correct'
solu.ons.
• Can
do
either.
Where
textual
traceability
(jus.fica.on)
is
essen.al,
knowledge
heavy
is
important.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
41
43. Overall
Approach
• Decompose
large
complex
problems
into
smaller,
manageable
problems
for
which
we
can
create
solu.ons.
• Soware
engineering
approach.
• Papers
by
Wyner
and
Peters
(2010,
2011).
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
42
44. Development
Caveat
• Developing
working
prototypes
(much
less
public
and/or
commercial
tools)
takes
resources.
• Tool
development
• Corpus
development
• Language
analysis
• It
is
a
slow,
painstaking,
and
gradual
process
of
construc.ng
modules
to
do
the
small
tasks
you
need
to
build
the
large
applica.ons
you
want.
• Not
a
simple
phone
app.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
43
45. Development
Cycle
Source
Text
Linguis.c
Analysis
Tool
Construc.on
Evalua.on
Knowledge
Extrac.on
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
44
48. Computa.onal
Linguis.c
Cascade
I
• Sentence
segmenta.on
-‐
divide
text
into
sentences.
• Tokenisa.on
-‐
words
iden.fied
by
spaces
between
them.
• Part
of
speech
tagging
-‐
noun,
verb,
adjec.ve....
• Morphological
analysis
-‐
singular/plural,
tense,
nominalisa.on,
...
• Shallow
syntac.c
parsing/chunking
-‐
noun
phrase,
verb
phrase,
subordinate
clause,
....
• Named
en.ty
recogni.on
-‐
the
en..es
in
the
text.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
47
49. Computa.onal
Linguis.c
Cascade
II
• Dependency
analysis
–
sentence
subject,
subordinate
clauses,
pronominal
anaphora,...
• Rela.onship
recogni.on
–
X
is
president
of
Y;
A
hit
B
with
a
car
and
killed
B.
• Enrichment
-‐
add
lexical
seman.c
informa.on
to
verbs
or
nouns.
• Supertagging
–
adding
conceptual
annota.ons
to
text.
• Transla.on
to
logic
for
reasoning.
• Each
step
guided
by
pa[ern
matching
and
rule
applica.on.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
48
50. Overall
Processing
Strategy
• Make
implicit
informa.on
explicit
by
adding
machine
readable
annota7ons.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
49
51. A
Tool
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
50
52. GATE
• General
Architecture
for
Text
Engineering
(GATE)
-‐
open
source
framework
which
supports
plug-‐in
NLP
components
to
process
a
corpus
of
text.
• GATE
Training
Courses
h[ps://gate.ac.uk/
• A
GUI
to
work
with
the
tools.
• A
Java
library
to
develop
further
applica.ons.
• Components
and
sequences
of
processes,
each
process
feeding
the
next
in
a
“pipeline”.
• Annotated
text
output
or
other
sorts
of
output.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
51
53. GATE
Benefits
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
52
• No
need
for
parsed,
pre-‐structured
text.
• Generic
components
apply
anywhere.
• No
need
for
a
gold
standard.
• Low
entry
point,
no
programming
required.
• Useful
interface
for
analysis
and
demonstra.on.
• Lots
of
public
resources
and
open
to
build
more
add-‐ons.
• Connects
to
other
tools,
widely
used....
54. GATE
Basic
Process
Flow
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
53
Can
add
further
processing
components
to
pipeline,
e.g.
NER,
co-‐reference,
other
other
annota.ons,...
55. GATE
-‐
Gaze[eers
• Gaze[eers
are
lookup
lists
that
add
features
-‐
when
a
string
in
the
text
is
located
in
a
lookup
list,
annotate
the
string
in
the
text
with
the
feature.
Conceptual
covers.
• Feature:
list
of
items...
• Obliga.on:
ought,
must,
obliged,
obliga.on....
• Excep.on:
unless,
except,
but,
apart
from....
• Verbs
according
to
thema.c
roles:
lists
of
verbs
and
their
associated
roles,
e.g.
run
has
an
agent
(Bill
ran),
rise
has
a
theme
(The
wind
blew).
07/09/2014
Argumenta.on
Summer
School,
Dundee
A.
Wyner,
Univ
of
Aberdeen
54
56. GATE
–
JAPE
Rules
• JAPE
Rules
(finite
state
transduc.on
rules)
create
overt
annota.ons
and
reuse
other
annota.ons
(e.g.
Parser
Output):
07/09/2014
Argumenta.on
Summer
School,
Dundee
A.
Wyner,
Univ
of
Aberdeen
55
57. GATE
–
Building
an
Applica.on
• Have
Gaze[eer
lists
and
JAPE
rules
for:
• lists
in
various
forms;
• excep.on
phrases
in
various
forms;
• condi.onals
in
various
forms;
• deon.c
terms;
• associa.ng
gramma.cal
roles
(e.g.
subject
and
object)
with
thema.c
roles
(agent
and
theme)
in
various
forms.
07/09/2014
Argumenta.on
Summer
School,
Dundee
A.
Wyner,
Univ
of
Aberdeen
56
58. Example
-‐
Camera
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
57
59. Argument
Fragment
for
a
Camera
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
58
60. Pro
and
Con
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
59
61. Comments
on
Comments
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
60
62. Goals
• Extract
arguments
distributed
across
a
corpora
and
evaluate
them
with
formal,
automated
tools.
• Speed
the
work
of
human
analysts.
• Provide
semi-‐automa3c
support.
• Use
aspects
of
NLP
to
incrementally
address
a
range
of
problems
(ambiguity,
structure,
contrasts,....)
• Wyner,
Schneider,
Atkinson,
and
Bench-‐Capon
(2012).
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
61
63. Consumer
Argumenta.on
Scheme
Variables
in
schemes
as
targets
for
extrac7on.
Premises:
• Camera
X
has
property
P.
• Property
P
promotes
value
V
for
agent
A.
Conclusion:
• Agent
A
should
Ac;on
Camera
X.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
62
64. Analyst’s
Goal:
Instan.ate
Premises:
• The
Canon
SX220
has
good
video
quality.
• Good
video
quality
promotes
image
quality
for
casual
photographers.
Conclusion:
• Casual
photographers
should
buy
the
Canon
SX220.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
63
65. Annota.ng
Text
• Annotate
text:
– Simple
or
complex
annota.ons.
– Highlight
annota.ons
with
– Search
for
and
extract
text
by
annota.on.
• GATE
“General
Architecture
for
Text
Engineering”.
– Works
with
large
corpora
of
text.
– Rule-‐based
or
machine-‐learning
approaches.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
64
66. To
Find
Argument
Passages
• Use:
– Indicators
of
aJer,
as,
because,
for,
since,
when,
....
– Indicators
of
therefore,
in
conclusion,
consequently,
....
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
65
68. To
Find
What
is
Being
Discussed
• Use
:
– Has
a
flash
– Number
of
megapixels
– Scope
of
the
zoom
– Lens
size
– The
warranty
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
67
70. To
Find
A[acks
Between
Arguments
• Use
contrast
terminology:
– Indicators
but,
except,
not,
never,
no,
....
– Contras.ng
sen.ment
The
flash
worked
.
The
flash
worked
.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
69
72. ,
,
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
71
73. Query
for
Pa[erns
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
72
74. An
Argument
for
Buying
the
Camera
Premises:
The
pictures
are
perfectly
exposed.
The
pictures
are
well-‐focused.
No
camera
shake.
Good
video
quality.
Each
of
these
proper.es
promotes
image
quality.
Conclusion:
(You,
the
reader,)
should
buy
the
CanonSX220.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
73
75. An
Argument
for
NOT
Buying
the
Camera
Premises:
The
colour
is
poor
when
using
the
flash.
The
images
are
not
crisp
when
using
the
flash.
The
flash
causes
a
shadow.
Each
of
these
proper.es
demotes
image
quality.
!
Conclusion:
(You,
the
reader,)
should
not
buy
the
CanonSX220.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
74
76. Counterarguments
to
the
Premises
of
“Don’t
buy”
The
colour
is
poor
when
using
the
flash.
For
good
colour,
use
the
colour
seZng,
not
the
flash.
The
images
are
not
crisp
when
using
the
flash.
No
need
to
use
flash
even
in
low
light.
The
flash
causes
a
shadow.
There
is
a
correc.ve
video
about
the
flash
shadow.
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
75
77. In
More
Detail
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
76
92. Example
-‐
Rules
• Rule
iden.fica.on
in
regula.ons;
what
one
can
'argue'
for
and
against.
• Using
previous
modules.
• Wyner
and
Peters
(2011)
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
91
93. Sample
Outputs
Consequence,
list
structure,
and
conjuncts
of
the
antecedent.
Excep.on,
agent
NP,
deon.c
concept,
ac.ve
main
verb,
theme.
07/09/2014
Argumenta.on
Summer
School,
Dundee
A.
Wyner,
Univ
of
Aberdeen
92
94. Sample
Output
Theme,
deon.c
modal,
passive
verb,
agent
with
complex
rela.ve
clause.
07/09/2014
Argumenta.on
Summer
School,
Dundee
A.
Wyner,
Univ
of
Aberdeen
93
95. Sample
Output
-‐
Overall
07/09/2014
Argumenta.on
Summer
School,
Dundee
A.
Wyner,
Univ
of
Aberdeen
94
96. Sample
Output
-‐
XML
07/09/2014
Argumenta.on
Summer
School,
Dundee
A.
Wyner,
Univ
of
Aberdeen
95
This
is
an
inline
representa.on,
and
not
'pure'
XML
as
tags
can
overlap.
There
is
also
offset,
which
can
be
modified
easily.
97. Sample
Output
–
ANNIC
Search
07/09/2014
Argumenta.on
Summer
School,
Dundee
A.
Wyner,
Univ
of
Aberdeen
96
99. Teamware
to
Create
Gold
Standards
07/09/2014
Argumenta.on
Summer
School,
Dundee
A.
Wyner,
Univ
of
Aberdeen
98
100. Results
of
Annota.on
• The
annotators
carry
out
their
task
and
complete
the
project.
• Carry
out
inter-‐annotator
agreement
analysis.
• Curate
the
disagreements
to
create
a
Gold
Standard
corpus.
Can
use
this
for
machine
learning.
• Search
the
annota.ons
using
an
online
tool,
e.g.
ANNIC.
07/09/2014
Argumenta.on
Summer
School,
Dundee
A.
Wyner,
Univ
of
Aberdeen
99
102. Add
to
Explorer
(or
Teamware)
• Verbs
for
proposi.onal
aZtudes,
e.g.
believe,
know,
hope
and
speech
acts,
e.g.
stated,
men7oned,
guessed.
• Opinion
adverbials
-‐
obviously,
so
far
as
I
know,
scien7fically.
• Ques.on
words
and
markers
–
who,
why,
?
• Rhetorical
connec.ves
-‐
elabora7on,
example,
contrast.
• Others....
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
101
103. References
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
• Wyner,
van
Engers,
Hunter
(2010)
• Wyner
and
Peters
(2010,
2011)
• Wyner,
Schneider,
Atkinson,
and
Bench-‐Capon
(2012)
102
104. Thanks
for
your
a[en.on!
• Questions?
• Contacts:
– Adam Wyner adam@wyner.info
Argumenta.on
Summer
School,
Dundee
07/09/2014
A.
Wyner,
Univ
of
Aberdeen
103