Comments on Natural Language and Argumentation

impulseverseAI and Robotics

Oct 24, 2013 (4 years and 15 days ago)

104 views

Comments on Natural Language
and Argumentation

Adam Wyner

Department
of Computer
Science





July 13, 2012

London Text Analytics
Meetup


Overview

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

2


Problem statement.


Representational layers:


Abstract argumentation.


Argumentation schemes.


Semi
-
automated argument analysis.


Well
-
formedness

of argumentation schemes.


Contrast
identification
.


Sketch the last three.

Problem

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

3


Arguments are everywhere.


Arguments are expressed in natural language.


Abstract arguments can be represented, related, and
reasoned with formally and computationally in
argumentation frameworks
.



Problem: How to get from arguments and contrasts
from a corpus of natural language into an abstract
representation in an argumentation framework?

Argument fragment for a camera

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

4

Pro and Con

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

5

Layers

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

6

Abstract argumentation

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

7

Input Graphs

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

8

http://rull.dbai.tuwien.ac.at:8080/ASPARTIX/
index.faces

Output Extensions

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

9

Preferred Extension

Argument ladder (
ArgMAS

2012)

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

10

Canonical sentences

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

11

Instantiation of the Position to Know Argumentation Scheme

Functional roles and typed propositional
functions

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

12

An abstract argument variable is functionally tied to the
propositions that represent the argumentation scheme, bridging
the representational levels.

Question

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

13



How
to
systematically associate
natural language
expressions
with an argumentation scheme
so as to
instantiate the
scheme, then use it for reasoning?


Manual Argument Analysis

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

14

C
oarse grained and uses no natural language processing.

Goals


Extract arguments from source texts so they can be
evaluated with formal automated tools.


Speed the work of human analysts.


Make argument identification more objective,
systematic, structured, and amenable to
development.


Manual
-
> Semi
-
automatic support
-
> More semi
-
automatic support
-
> Fully automatic.


Use aspects of NLP to incrementally address a range
of problems (ambiguity, structure, contrasts,....)

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

15

Strategy and issues


Decompose the complexity of a text


What are the parts of an
argument
?


How
are the parts of the argument related
?


What are the 'boundaries' of an argument?


What are the contrasts and negations from which
we can derive attack relationships?


What kind of domain knowledge do we need?


Take a rule
-
based approach.


Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

16

Use case:

Which camera should I buy?

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

17

Value
-
based Practical Reasoning
Argumentation Scheme

Premises:


Before doing action A, the current circumstances are R;


After doing action A, the new circumstances are S;


G is a goal of the agent Ag, where S implies G;


Doing action A in R and achieving G promotes value V;


Conclusion:


We should perform action A.


Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

18

Consumer Argumentation Scheme

Premises:


Camera X has property P.


Property P promotes value V for agent A.


Conclusion:


Agent A should
Action1

Camera X.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

19

Critical questions


Does Camera X have property P?


Does property P promote value V for agent A?


Is value V more important than value V’ for agent A?


Answers can let presumptive conclusion remain or be
challenged.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

20

Analyst’s goal: instantiate

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
.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

21

… starting from this

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

22

Highlight parts of the argument


Camera X has
property

P.


P
roperty P promotes
value

V for
agent

A.


V
alue V is
more important

than value V’ for agent A.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

23

To find and instantiate the argument


Argumentative indicators


Property


with camera terminology


Value

for
agent



with sentiment, user models


Value V
more important



with comparisons

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

24

Implementation with GATE


GATE “General Architecture for Text Engineering”.


Environment for text analysis.


Manual, semi
-
automatic, fully automatic.


Adds annotation to text:


Can work with large corpora of text


Coarse or fine
-
grained annotations


Rule
-
based or machine
-
learning
.


Highlight annotations with


Search for and extract annotated text.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

25

To find argument passages


Use:


Indicators of


after, as, because, for, since, when, ....


Indicators of


therefore, in conclusion, consequently, ....

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

26

Rhetorical terminology

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

27

To find what is being discussed


Use domain terminology:


Has a flash


Number of megapixels


Scope of the zoom


Lens size


The warranty

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

28

Domain terminology

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

29

To find attacks between arguments


Use contrast terminology:


Indicators


but, except, not, never, no, ....


Contrasting sentiment


The
flash worked
.

The
flash worked
.


Other contrast issues later.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

30

Sentiment terminology

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

31

Agents: user
m
odels


User’s parameters


Age, gender, education, previous camera experience, ....


User’s context of use


Party, indoors, sport, travel, desired output format, ....


User’s constraints


Cost, portability, size, richness or flexibility of features, ....


User’s quality expectations


Colour

quality, information density, reliability, ....

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

32

Instantiating the CAS

Premises:



The Canon SX220 camera has property P.


Property P promotes value V for agent A.


Conclusion:



Agent A should buy the Canon SX220.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

33

,

,

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

34

Query for patterns

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

35

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 properties promotes image quality.


Conclusion:



(You, the reader,) should buy the CanonSX220.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

36

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 properties demotes image quality.


Conclusion:



(You, the reader,) should not

buy the CanonSX220.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

37

Counterarguments to the premises of
“Don’t buy”




The
colour

is poor when using the flash.



For good
colour
, use the
colour

setting, 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 corrective video about the flash shadow.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

38

Locating argumentation schemes from text


What is a well
-
formed argumentation scheme? Need
to know in order to have some idea what textual
indicators to look for in a corpus. An open question.


Steps to address it (CMN 2012).


Narrative coherence



rhetorical indicators,
sentiment, negation, tense/aspect, roles,....


Corpus to work with.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

39

Preliminary work

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

40

How are contrasting pairs to be identified?


Given a sentence and a corpus, find contrasting sentences
.


Compare sentences for textual similarity.


Identify textual contrasts


negation, antonyms.


The value of budget is promoted.


The value of budget is not promoted.


The value of budget is demoted.


Address diathesis, e.g. active and passive sentence forms


Bill returned the book.


The book was returned by Bill.


The book was not returned by Bill


Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

41

How are contrasting pairs to be identified?


Similarity measure (list comparison between
sentences) using not just the text itself but also
annotations for parts of speech and grammatical
phrases.


Find contrast indicators, e.g. ''not'', and tag for
antonyms.


Issues


scope, scale up, relate to similar work on
textual inference and contradiction.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

42

Knowledge light v. heavy
a
pproaches


Knowledge light in terms of knowledge of the domain or of
language


statistical or machine learning approaches.
Algorithmically compare and contrast large bodies of textual
data, identifying regularities and similarities. Sparse data
problem. Need a gold standard. No
rules extracted
. Opaque.


Knowledge heavy
-

lists, rules, and processes. Labour and
knowledge intensive. Transparent. Reasoning to annotation.


Can do either. Depends what one wants. Finding what one
knows in sparse data v. finding unknowns in rich data.

13/7/12

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

43

Future work


Tool refinement.


Add domain and ontology modules to the tool.


User models


how do they play a role?


More complicated query patterns, what results do we get?


More elaborate examples.


Disambiguation issues for rhetorical terminology, e.g.
when
,
because
,.... Deal with it step
-
by
-
step to find how to
disambiguate the indicators or other terminology.


Further work on argumentation scheme characterisation.


Further work on contrariness.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

44

Acknowledgements


FP7
-
ICT
-
2009
-
4
Programme
, IMPACT Project,

Grant Agreement Number 247228.


Collaborators: Jodi Schneider, Trevor Bench
-
Capon, Katie
Atkinson, and
Chenhui

Lui
.

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

45

Thanks for your attention!


Questions?


Contacts:



Adam Wyner


adam@wyner.info

Wyner, Text Analytics Meetup, London, UK,
(cc) by
-
nc
-
sa license

13/7/12

46