Finding Counter-Arguments

Computational Argumentation WS 2019/2020

Masa Maric
04.11.2019
LMU CIS

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 1 / 34

Introduction

Argumentation is an appealing way to reason from defective premises.

Various approaches have been proposed as computational models for

argumentation. Most of them deal with e.g. establishing arguments for or
against a conclusion, identifying counter-arguments, defining an attack
relation between arguments, etc.

Generating arguments - an argumentative model where arguments are pairs

(premises,conclusion) such that conclusion is a logical formula classically
entailed by premises, a nonordered collection of logical formulae

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 2 / 34

Argumentation

Propositional (statement) logic

Logical formulae are denoted α,β,γ,... and sets thereof are denoted Φ,Ψ,Θ.

knowledge base ∆ consisting of a finite set of clauses

we assume a canonical enumeration for its every subset

Next: Besnar and Hunter argumentative model: an argument is a set of

formulae entailing another formula viewed as the conclusion of the
argument.

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 3 / 34

Besnar and Hunter

Definition 1: An argument A is a pair (Φ,α) if:

The sets Φ and α denote the support and the conclusion of A,

respectively.

Definition 2: An argument (Φ,α) is more conservative than an argument

(Ψ,β) if

Definition 3:

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 4 / 34

Besnar and Hunter, ctd.

Definition 4:

Definition 5:

Definition 6:

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 5 / 34

Besnar and Hunter, ctd.

Definition 7:

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 6 / 34

Minimal Unsatisfiable Set (MUS)

MUS is an unsatisfiable set of formulae which becomes satisfiable if

deprived of any member; all proper subsets of a MUS are satisfiable
although it is itself unsatisfiable.

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 7 / 34

Generating Counter-Arguments

argumentation involves a number of arguments and counter-arguments

for/against some conclusion

Generating an argument for this conclusion requires extracting a minimal

consistent subset entailing this conclusion

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 8 / 34

Generating Counter-Arguments

Proposition:

Proposition 2 suggests that computing arguments is equivalent with

computing MUSes so that algorithms that often prove efficient in finding
MUSes can be taken advantage of

However, such algorithms do not work with arbitrary formulae but clauses

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 9 / 34

Generating Counter-Arguments

Theorem:

Due to argumentation trees, a case of utmost importance is that of

computing canonical undercuts. Dealing with these turns out to be simpler
because only MUSes of ∆ need be considered

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 10 / 34

Algorithms

1. algorithm – BA

- generates all arguments for a specific conclusion

- first converts NOT α into clausal form

- next step consists in extracting the MUSes from ∆∪{¬α} that

contain at least one clause of ¬α

- HYCAM algorithm (Hybridization for Computing All Muses); finds

MUSes through Maximal Satfisfiable Subsets

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 11 / 34

Algorithm

2. algorithm – BT

- generates the complete argumentation tree for a given argument which

is the root of the tree

-other nodes are recorded in a stack as they are found, popping the last
item as its canonical undercuts are obtained

- requires the counter-argument being computed to have its

support to include some new piece of information thus precluding any
cycles

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 12 / 34

 Literature
MUS-Based Generation of Arguments and Counter-arguments -
https://www.researchgate.net/publication/220712497_MUS-Based_Generation_of_
ents_and_Counter-arguments

Exploiting Background Knowledge for Argumentative Relation Classification -

http://drops.dagstuhl.de/opus/volltexte/2019/10372/pdf/OASIcs-LDK-2019-8.pdf
Propositional Logic - http://infolab.stanford.edu/~ullman/focs/ch12.pdf
Enumerated Sets -
http://doc.sagemath.org/html/en/reference/categories/sage/categories/enumerate
html
MUS - http://www.epcl-study.eu/content/downloads/slides/belov_2012_lecture.pdf
Algorithms for generating arguments and counterarguments in propositional logic -
http://csri.gr/files/publications/3.algorithms%20for%20generating%20arguments%2
%20counterarguments%20in%20classical%20propositional%20logic.pdf
LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 13 / 34
 THANK YOU FOR YOUR
ATTENTION!

LMU Department of Media Informatics | Proseminar Medieninformatik WS 2011/2012 | email@aendern.de Slide 14 / 34