Chapter 16: Rationality Postulates and Critical Examples Martin - - PowerPoint PPT Presentation
Chapter 16: Rationality Postulates and Critical Examples Martin - - PowerPoint PPT Presentation
Chapter 16: Rationality Postulates and Critical Examples Martin Caminada Department of Computing Science University of Aberdeen Outline (1) introduction (2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted
Outline
(1) introduction
(2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted rebut solutions b) unrestricted rebut solutions (4) non-interference and crash resistance a) erasing inconsistent arguments b) requiring consistent entailment and forbidding strict-on-strict (5) rationality postulates and other instantiations (6) summary and discussion
Outline
(1) introduction
(2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted rebut solutions b) unrestricted rebut solutions (4) non-interference and crash resistance a) erasing inconsistent arguments b) requiring consistent entailment and forbidding strict-on-strict (5) rationality postulates and other instantiations (6) summary and discussion
Introduction
- explain difference between strict
and defeasible inference steps
- problems start when combining these;
this is often required (explain)
- in argumentation context, part of the problem
is applying blind semantics (explain)
- satisfying a reasonable outcome is far from trivial;
in current chapter we try to specify what a reasonable
- utcome is, and how to bring this about
Outline
(1) introduction
(2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted rebut solutions b) unrestricted rebut solutions (4) non-interference and crash resistance a) erasing inconsistent arguments b) requiring consistent entailment and forbidding strict-on-strict (5) rationality postulates and other instantiations (6) summary and discussion
Preliminaries
- show how to construct arguments
using strict and defeasible rules (simplified ASPIC)
- introduce some ways
- f dealing with preferences:
elitist vs democratic
- distinguish restricted rebut
from unrestricted rebut
Outline
(1) introduction
(2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted rebut solutions b) unrestricted rebut solutions (4) non-interference and crash resistance a) erasing inconsistent arguments b) requiring consistent entailment and forbidding strict-on-strict (5) rationality postulates and other instantiations (6) summary and discussion
Direct Consistency, Indirect Consistency and Closure
- give Married John example
from (Caminada & Amgoud 2007)
- informally explain that under unrestricted rebut,
the example violates closure and indirect consistency
- informally explain that under restricted rebut,
the example violates direct consistency
- now provide formal definitions of postulates
direct consistency, indirect consistency and closure
- mention that two classes of solutions have been found:
- ne based on restricted rebut,
- ne based on unrestricted rebut
Outline
(1) introduction
(2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted rebut solutions b) unrestricted rebut solutions (4) non-interference and crash resistance a) erasing inconsistent arguments b) requiring consistent entailment and forbidding strict-on-strict (5) rationality postulates and other instantiations (6) summary and discussion
Restricted Rebut Solutions
- i. transposition
- ii. contraposition
- iii. semi-abstract approach of Dung&Tang
- iv. on the need of complete-based semantics
Restricted Rebut Solutions
- i. transposition
- show that Married John example will be OK
- provide formal definition of transposition
- why do we need restr. rebut? Show tandem example.
problem AA: attack is binary but conflict can be ternary
- Theorem:
restr rebut + complete-based sem + transposition = postulates satisfied
- ii. contraposition
- iii. semi-abstract approach of Dung&Tang
- iv. on the need of complete-based semantics
Restricted Rebut Solutions
- i. transposition
- ii. contraposition
- contraposition is an alternative to transposition
- provide formal definition of contraposition
in the context of strict rules
- Theorem:
restr rebut + complete-based sem + contraposition = postulates satisfied
- iii. semi-abstract approach of Dung&Tang
- iv. on the need of complete-based semantics
Restricted Rebut Solutions
- i. transposition
- ii. contraposition
- iii. semi-abstract approach of Dung&Tang
- D&T provide semi-abstract approach
based on subargument structure and attack relation
- provide formal definitions
- D&T claim their solution is very general:
transposition and contraposition are special cases
- however, only works for restricted rebut (!)
- iv. on the need of complete-based semantics
Restricted Rebut Solutions
- i. transposition
- ii. contraposition
- iii. semi-abstract approach of Dung&Tang
- iv. on the need of complete-based semantics
- the above 3 approaches rely on complete-based sem
- things start to fail when applying non-adm-based sem
- provide counter example against naive semantics
- provide counter example against stage semantics
- provide counter example against CF2 semantics
- it is not clear how non-adm-based semantics
can be used for any meaningful instantiations
Outline
(1) introduction
(2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted rebut solutions b) unrestricted rebut solutions (4) non-interference and crash resistance a) erasing inconsistent arguments b) requiring consistent entailment and forbidding strict-on-strict (5) rationality postulates and other instantiations (6) summary and discussion
Unrestricted Rebut Solutions
- provide example COMMA 2014 to illustrate
shortcomings of unrestricted rebut
- shortcomings are specially bad in dialogue context
- for unrestricted rebut, complete-based semantics is not
sufficient (tandem example) grounded is really needed
- Theorem:
unrestricted rebut + grounded sem + transposition = postulates satisfied
- works:
- when preferences are empty (C&A 2007)
- when preferences are total pre-order (CMO 2014)
- restricted rebut vs unrestricted rebut:
“every advantage has its disadvantage”
Outline
(1) introduction
(2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted rebut solutions b) unrestricted rebut solutions (4) non-interference and crash resistance a) erasing inconsistent arguments b) requiring consistent entailment and forbidding strict-on-strict (5) rationality postulates and other instantiations (6) summary and discussion
Non-Interference and Crash-Resistance
- why not use classical logic to generate the strict rules?
transposition would come for free, so easy way of satisfying postulates!
- unfortunately: new problem caused by ex falso quodlibet
- provide coffee example (BNAIC 2005)
- Pollock, Reiter and ASPIC+ tried to solve this
by applying preferred or stable semantics
- this still doesn't solve things;
provide unrel John & unrel Mary example (BNAIC 2005)
- formally define the additional rationality postulates of
non-interference and crash-resistance
- explain that semantics alone cannot solve the problem;
something needs to change on how to construct the graph
Outline
(1) introduction
(2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted rebut solutions b) unrestricted rebut solutions (4) non-interference and crash resistance a) erasing inconsistent arguments b) requiring consistent entailment and forbidding strict-on-strict (5) rationality postulates and other instantiations (6) summary and discussion
Erasing Inconsistent Arguments
- why not just remove inconsistent arguments,
as these are clearly absurd
- warning: removing “absurd” arguments can lead to
problems; self-undercutting arguments cannot be removed (see Pollock's pink elephant example)
- we need to show that removing inconsistent arguments
doesn't lead to similar problems; this is far from trivial!
- Wu&Podlaszewski proved that removing inconsistent
arguments satisfies all rationality postulates! (Theorem)
- their solution only works with restricted rebut,
and when preferences are empty (counterexample Leon)
Outline
(1) introduction
(2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted rebut solutions b) unrestricted rebut solutions (4) non-interference and crash resistance a) erasing inconsistent arguments b) requiring consistent entailment and forbidding strict-on-strict (5) rationality postulates and other instantiations (6) summary and discussion
Requiring Consistent Entailment and Forbidding Strict-on-Strict
- idea of Grooters & Prakken COMMA 2014
- change the way in which strict rules
are generated by classical logic (consistent entailment only)
- no strict rule can feed into another strict rule
(have to have defeasible rule in between)
- is not just an instantiation of ASPIC+
instead it defines a whole different ASPIC version!
- Theorem: all postulates satisfied (are they?)
- advantage: it works with preferences (unlike W&P)
- disadvantage: works only with restricted rebut
Outline
(1) introduction
(2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted rebut solutions b) unrestricted rebut solutions (4) non-interference and crash resistance a) erasing inconsistent arguments b) requiring consistent entailment and forbidding strict-on-strict (5) rationality postulates and other instantiations (6) summary and discussion
Rationality Postulates and Other Instantiations
- so far, we have only studied rule-based instantiations
- however, similar problems also occur
in classical logic-based instantiations
- restricted rebut and unrestricted rebut
have equivalents in logic-based instantiations (Gorogiannis & Hunter 2011)
- explain how tandem example works
in Toni Hunter's approach; same problems and space of solution
Outline
(1) introduction
(2) preliminaries (3) direct consistency, indirect consistency and closure a) restricted rebut solutions b) unrestricted rebut solutions (4) non-interference and crash resistance a) erasing inconsistent arguments b) requiring consistent entailment and forbidding strict-on-strict (5) rationality postulates and other instantiations (6) summary and discussion
Summary and Discussion
- (optionally) provide a table with some of the key results
- most papers are on abstract level
most problems are on instantiated level
- provide “care home analogy”
to illustrate problem of “blind argument selection”
- explain that postulates are not just for logical elegance;
we need these to take into account real world constraints
- non-admissibility based semantics are nice on
the abstract level; not so nice on the non-abstract level
- if the problem is constraint satisfaction, why not use CAFs?
CAFs don't work on any of the examples! abstract “solutions” make things worse instead of better!
- what is it that abstract theories are abstractions of?
Example
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
Example
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st
Example
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
Example
A4 A5 A6 A1 A2 A3
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
Example
A4 A5 A6 A1 A2 A3
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
Example
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
jt mt st ¬jt ¬mt ¬st A4 A5 A6 A1 A2 A3
Example
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
jt mt st ¬jt ¬mt ¬st A4 A5 A6 A1 A2 A3
Example
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
jt mt st ¬jt ¬mt ¬st A4 A5 A6 A1 A2 A3
Example
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
jt mt st ¬jt ¬mt ¬st A4 A5 A6 A1 A2 A3
Example
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
A4 A5 A6 A1 A2 A3
Example
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
A4 A5 A6 A1 A2 A3
Example
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
A4 A5 A6 A1 A2 A3
Example
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt; mw ⇒ mt; sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
jt mt st ¬jt ¬mt ¬st A4 A5 A6 A1 A2 A3
Rationality Postulates
Let J be the set of justified conclusions and ClS(J) be the closure of J under the strict rules in S. direct consistency: ¬∃ p: (p ∈ J ∧ ¬p ∈ J) closure: J = ClS(J) indirect consistency: ¬∃ p: (p ∈ClS(J) ∧ ¬p ∈ ClS(J))
Satisfying the Rationality Postulates
(1) use restricted rebut and any complete-based semantics (+ special condition on the strict rules) (2) use unrestricted rebut and grounded semantics (+ special condition on the strict rules)
Common Mistakes
Some common mistakes in argumentation: 1) not applying transposition 2) not applying a complete-based semantics 3) constructing strict rules from classical logic 4) messing around with the graph
Common Mistakes
Some common mistakes in argumentation: 1) not applying transposition 2) not applying a complete-based semantics 3) constructing strict rules from classical logic 4) messing around with the graph
Transposition
Take the following strict rule: a1, ..., ai-1, ai, ai+1, ..., an c A transposition of this rule is: a1, ..., ai-1, ¬c, ai+1, ..., an ¬ai (for some 1 ≤ i ≤ n) A set of strict rules S is closed under transposition iff it contains all transpositions of the rules in S.
Without Transposition
S = { r; p; m hs; b ¬hs } D = { r ⇒m; p ⇒ b } A1: (r) ⇒ m A2: (p) ⇒ b A3: A1 hs A4: A2 ¬hs
With Transposition
S = { r; p; m hs; b ¬hs ¬hs ¬m; hs ¬b} D = { r ⇒m; p ⇒ b } A1: (r) ⇒ m A2: (p) ⇒ b A3: A1 hs A4: A2 ¬hs
With Transposition
S = { r; p; m hs; b ¬hs ¬hs ¬m; hs ¬b} D = { r ⇒m; p ⇒ b } A1: (r) ⇒ m A2: (p) ⇒ b A3: A1 hs A4: A2 ¬hs A5: A3 ¬b A6: A4 ¬m
With Transposition
S = { r; p; m hs; b ¬hs ¬hs ¬m; hs ¬b} D = { r ⇒m; p ⇒ b } A1: (r) ⇒ m Lack of transposition A2: (p) ⇒ b causes problems in A3: A1 hs
- Prakken & Sartor 1997
A4: A2 ¬hs
- DeLP
A5: A3 ¬b
- Nute's Defeasible Logic
A6: A4 ¬m
Common Mistakes
Some common mistakes in argumentation: 1) not applying transposition 2) not applying a complete-based semantics 3) constructing strict rules from classical logic 4) messing around with the graph
Common Mistakes
Some common mistakes in argumentation: 1) not applying transposition 2) not applying a complete-based semantics 3) constructing strict rules from classical logic 4) messing around with the graph
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt, mw ⇒ mt, sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
A4 A5 A6 A1 A2 A3
Naive Semantics
Naive Semantics
A1 A2 A3 A4 A5 A6
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt, mw ⇒ mt, sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
Naive Semantics
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt, mw ⇒ mt, sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
jt mt st A1 A2 A3 A4 A5 A6
Stage Semantics
A1 A2 A3 A4 A5 A6
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt, mw ⇒ mt, sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
Stage Semantics
A1 A2 A3 A4 A5 A6
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt, mw ⇒ mt, sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
Stage Semantics
A1 A2 A3 A4 A5 A6
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt, mw ⇒ mt, sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
Semantics and Consistency
Semantics that gets it right:
- complete
- semi-stable
- grounded
- ideal
- preferred
- eager
Semantics that get it wrong:
- naïve
- CF2
- stage
- stage2
Common Mistakes
Some common mistakes in argumentation: 1) not applying transposition 2) not applying a complete-based semantics 3) constructing strict rules from classical logic 4) messing around with the graph
Common Mistakes
Some common mistakes in argumentation: 1) not applying transposition 2) not applying a complete-based semantics 3) constructing strict rules from classical logic 4) messing around with the graph
Strict Rules as Logical Inferences
P = { r; p; m ⊃ hs; b ⊃ ¬hs } (“”≡ “⊢”) D = { r ⇒ m; p ⇒ b } A1: (r) ⇒ m A2: (p) ⇒ b A3: (A1, m ⊃ hs) hs A4: (A2, b ⊃ ¬hs) ¬hs A5: (A3, b ⊃ ¬hs) ¬b A6: (A4, m ⊃ hs) ¬m
Strict Rules as Logical Inferences
P = { j; m; wf } (“”≡ “⊢”) D = { j s; m ¬s; wf r } There now exist the following arguments: A = (j) s B = (m) ¬s D = (wf) r
Strict Rules as Logical Inferences
P = { j; m; wf } (“”≡ “⊢”) D = { j s; m ¬s; wf r } There now exist the following arguments: A = (j) s (unfortunately, B = (m) ¬s there also exists: D = (wf) r C = A, B ¬r)
Strict Rules as Logical Inferences
- Grounded semantics: no justified argu-
ments
- Why not use preferred or stable semantics?
- Reiter and Pollock also do this...
Strict Rules as Logical Inferences
John: “Cup of coffee contains sugar.” Mary: “Cup of coffee doesn't contain sugar.” John: “I'm unreliable.” Mary: “I'm unreliable.” Weather Forecaster: “Tomorrow rain.”
Strict Rules as Logical Inferences
Work that gets it wrong:
- ASPIC+
- Reiter's Default Logic
- pretty much everything of John Pollock
Work that gets it right:
- Gorogiannis & Hunter (AIJ 2011)
- Wu & Podlaszewski (JLC 2014)
- Prakken and Grooters (COMMA 2014)
- more work that is yet to come...
Common Mistakes
Some common mistakes in argumentation: 1) not applying transposition 2) not applying a complete-based semantics 3) constructing strict rules from classical logic 4) messing around with the graph
Common Mistakes
Some common mistakes in argumentation: 1) not applying transposition 2) not applying a complete-based semantics 3) constructing strict rules from classical logic 4) messing around with the graph
Value-Based Argumentation
A1 A2 A3 A4 A5 A6
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt, mw ⇒ mt, sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
Value-Based Argumentation
A1 A2 A3 A4 A5 A6
S = { jw; mw; sw; mt,st¬jt; jt,st¬mt; jt,mt¬st } D = { jw ⇒ jt, mw ⇒ mt, sw ⇒ st }
A1: (jw) ⇒ jt A2: (mw) ⇒ mt A3: (sw) ⇒ st A4: A2, A3 ¬jt A5: A1, A3 ¬mt A6: A1, A2 ¬st
“Enhanced” Argumentation Frame- works
Work that gets it wrong
- PAFs
- CAFs
- VAFs
- TAFs
- …
- …
Be warned: almost all “enhancements” of Dung's AFs have these kind of problems Work that gets it right:
- EAFs
- enhancements that are explicitly designed
for satisfying the rationality postulates
Common Mistakes
Some common mistakes in argumentation: 1) not applying transposition 2) not applying a complete-based semantics 3) constructing strict rules from classical logic 4) messing around with the graph
Take Home Message (1/3)
looks OK at the abstract level
not necessarily OK at the non-abstract level
Take Home Message (2/3) don't do research purely on the abstract level
Take Home Message (2/3) don't do research purely on the abstract level
unless you know it can be applied
- n the non-abstract level
Take Home Message (3/3)
If you do want to invent a new “abstraction” then please give at least one fully instantiated system and show that it satisfies reasonable properties
Research Challenge: finding the magic combinations
extensions (labellings)
- f conclusions
(3) determining status of conclusions
extensions (labellings)
- f arguments
(2) applying argumentation semantics
argumentation framework
(1) argument (+attack) construction
knowledge base
Rationality Postulates
Let J be a set of conclusions yielded by an argumentation formalism.
- direct consistency
J does not contain contraries (p and ¬p)
- closure
J is closed under the strict rules
- indirect consistency
the closure of J under strict rules is directly consistent
- crash-resistance
no set of formulas can make a totally unrelated set of formulas completely irrelevant, when being merged to it
- non-interference
no set of formulas can influence the entailment of a totally unrelated set of formulas, when being merged to it
Rationality Postulates
direct consistency closure indirect consistency crash-resistance non-interference backwards compatibility
Caminada & Amgoud AIJ 2007 Caminada, Dunne & Carnielli JLC 2011
Further Reading
- Abstract argumentation
- Dung, AIJ 1995 (landmark paper)
- Baroni, Caminada & Giacomin, KER 2011
- Instantiated argumentation & rationality postulates
- Caminada & Amgoud, AIJ 2007 (ASPIC)
- Modgil & Prakken, AIJ 2013 (ASPIC+)
- Gorogiannis & Hunter, AIJ 2011
- Wu & Podlaszewski, JLC 2014
- Caminada, Modgil & Oren, COMMA 2014
(ASPIC-)
Epilogue
- Why use a formalism based on
strict and defeasible rules?
- Why not just use classical logic, like
Gorogiannis & Hunter (AIJ 2011)
- Why not represent a defeasible rule
a, b, c ⇒ d as a material implication a ∧ b ∧ c d
Example
S = { → s; → p } D = { s ⇒ m; m ⇒ f; m ⇒ r; p ⇒ ¬r }
Example
S = { → s; → p } D = { s ⇒ m; m ⇒ f; m ⇒ r; p ⇒ ¬r }
Example
Example
Other Example
S = { → o; → a } D = { o ⇒ O(s); a ⇒ c; c ⇒ O(¬s) }
Other Example
S = { → o; → a } D = { o ⇒ O(s); a ⇒ c; c ⇒ O(¬s) } → o ⇒ O(s) → a ⇒ c ⇒ O(¬s)
On the Nature of Reasoning
- epistemic vs. constitutive reasoning
On the Nature of Reasoning
- epistemic vs. constitutive reasoning
- hard conflicts vs. soft conflicts
On the Nature of Reasoning
- epistemic vs. constitutive reasoning
- hard conflicts vs. soft conflicts
p q; ¬q
- ¬p