Tutorial: Nonmonotonic Logic (Day 1) Christian Straßer Institute for Philosophy II, Ruhr-University Bochum Center for Logic and Philosophy of Science, Ghent University http://homepage.ruhr-uni-bochum.de/defeasible-reasoning/index.html
September 2, 2015
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
1 / 104
Outline 1
2 3
4 5
Defeasible Reasoning – Some Basic Concepts Some examples to warm up Some conceptual distinctions Nonmonotonic Logic in Context The Dynamics of Defeasible Reasoning Default Logic (in the tradition of Reiter) Warming up Defaults and Default Theories Inferring Alternatives and more examples Meta-Properties Introducing Priorities Summing up Autoepistemic Logic Bibliography Bibliography
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
2 / 104
Overview: the whole tutorial
Day 1 1
2
motivation and basic concepts default logic (Reiter, Poole, Horty) Reiter and variants many examples priorities (Horty)
3
autoepistemic logic (Moore, Konolige, Marek) basics
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
3 / 104
Overview: the whole tutorial Day 2 Day 1 1
2
1
motivation and basic concepts default logic (Reiter, Poole, Horty)
default assumptions, adaptive logics reasoning with maximal consistent subsets bridging to preferential semantics meta-theory: interesting properties for non-monotonic logic
Reiter and variants many examples priorities (Horty) 3
autoepistemic logic (Moore, Konolige, Marek) basics
Christian Straßer (RUB, UGENT)
Plausible Reasoning (Makinson, Rescher/Manor, Batens, Geffner/Pearl)
2
Preferential Semantics (Kraus/Lehmann/Magidor, Shoham, Batens) properties, limitations, variants enhancements (rational closure)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
3 / 104
Overview: the whole tutorial Day 2 Day 1 1
2
1
motivation and basic concepts default logic (Reiter, Poole, Horty)
default assumptions, adaptive logics reasoning with maximal consistent subsets bridging to preferential semantics meta-theory: interesting properties for non-monotonic logic
Reiter and variants many examples priorities (Horty) 3
autoepistemic logic (Moore, Konolige, Marek)
Plausible Reasoning (Makinson, Rescher/Manor, Batens, Geffner/Pearl)
2
basics
Preferential Semantics (Kraus/Lehmann/Magidor, Shoham, Batens) properties, limitations, variants enhancements (rational closure)
Yes, . . . I know: it’s not realistic! Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
3 / 104
Golden Rule
interrupt me — ask — complain — comment
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
4 / 104
Slides
can be found at http://homepage.ruhr-uni-bochum.de/defeasible-reasoning/ Courses/Natal2015-NonMonLog/natal15-nonmonlog.html (already outdated . . . )
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
5 / 104
Topic 1
2 3
4 5
Defeasible Reasoning – Some Basic Concepts Some examples to warm up Some conceptual distinctions Nonmonotonic Logic in Context The Dynamics of Defeasible Reasoning Default Logic (in the tradition of Reiter) Warming up Defaults and Default Theories Inferring Alternatives and more examples Meta-Properties Introducing Priorities Summing up Autoepistemic Logic Bibliography Bibliography
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
6 / 104
What’s defeasible reasoning: some examples
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
7 / 104
What’s defeasible reasoning: some examples
abductive inference infer a (good!?) explanation α → β and β thus α Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
7 / 104
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
8 / 104
closed world assumption reasoning on the assumption that the given information is complete Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
8 / 104
Tweety is a bird. Thus, . . . ?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
9 / 104
Tweety is a bird. Thus, . . . ? stereotypical / default reasoning jump to a conclusion on the basis of what is usually/typically/normally/etc. the case
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
9 / 104
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
10 / 104
inductive generalisations
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
10 / 104
Domains of defeasible reasoning
everyday reasoning expert reasoning (e.g. medical diagnosis) scientific reasoning
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
11 / 104
Commonalities
tentative conclusions → jumping to conclusions retraction possible if problems arise
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
12 / 104
Commonalities
tentative conclusions → jumping to conclusions retraction possible if problems arise
Two tiers of defeasible reasoning 1
illative tier (support, concluding)
2
dialectic tier (retraction)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
12 / 104
Ampliative vs. Corrective approaches
Corrective approaches In contrast to ampliative reasoning, each inference is in accordance with CL (or another deductive standard) and hence deductive. However, given an inconsistent theory, not all deductive inferences will be accepted. illative tier: strictly deductive (e.g., classical logic) dialectic tier: conflicting deductive inferences
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
13 / 104
Ampliative vs. Corrective approaches
Corrective approaches In contrast to ampliative reasoning, each inference is in accordance with CL (or another deductive standard) and hence deductive. However, given an inconsistent theory, not all deductive inferences will be accepted. illative tier: strictly deductive (e.g., classical logic) dialectic tier: conflicting deductive inferences
Examples nonmonotonic paraconsistent logics (Rescher/Manor, inconsistency-adaptive logics) deductive argumentation-based approaches
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
13 / 104
Example: Rescher and Manor’s Free Consequences suppose we reason classically on the basis of a complex body of premises Γ (e.g., a mathematical or scientific theory, or code of law)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
14 / 104
Example: Rescher and Manor’s Free Consequences suppose we reason classically on the basis of a complex body of premises Γ (e.g., a mathematical or scientific theory, or code of law) we do not know whether Γ is consistent
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
14 / 104
Example: Rescher and Manor’s Free Consequences suppose we reason classically on the basis of a complex body of premises Γ (e.g., a mathematical or scientific theory, or code of law) we do not know whether Γ is consistent careful rationale for drawing inferences
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
14 / 104
Example: Rescher and Manor’s Free Consequences suppose we reason classically on the basis of a complex body of premises Γ (e.g., a mathematical or scientific theory, or code of law) we do not know whether Γ is consistent careful rationale for drawing inferences call a formula ϕ in Γ free if it does not belong to a minimally inconsistent set
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
14 / 104
Example: Rescher and Manor’s Free Consequences suppose we reason classically on the basis of a complex body of premises Γ (e.g., a mathematical or scientific theory, or code of law) we do not know whether Γ is consistent careful rationale for drawing inferences call a formula ϕ in Γ free if it does not belong to a minimally inconsistent set rationale: an inference in CL is retracted as soon as we find out that it relies on premises that are not free
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
14 / 104
Example: Rescher and Manor’s Free Consequences suppose we reason classically on the basis of a complex body of premises Γ (e.g., a mathematical or scientific theory, or code of law) we do not know whether Γ is consistent careful rationale for drawing inferences call a formula ϕ in Γ free if it does not belong to a minimally inconsistent set rationale: an inference in CL is retracted as soon as we find out that it relies on premises that are not free
Task: Free consequences What are the free consequences of Γ = {p ∧ q, ¬p, r ∧ s}?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
14 / 104
Ampliative approaches While the truth of the premises does not warrant the truth of the conclusion as in deductive reasoning, the conclusion nevertheless holds in most/typical/etc. cases in which the premises hold.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
15 / 104
Ampliative approaches While the truth of the premises does not warrant the truth of the conclusion as in deductive reasoning, the conclusion nevertheless holds in most/typical/etc. cases in which the premises hold. illative tier: beyond truth-preservation typically: fixed minimal deductive (non-defeasible) standard of reasoning / core rules (e.g., classical logic) jump to more conclusions given additional warrants that allow for defeasible conclusions
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
15 / 104
Ampliative approaches While the truth of the premises does not warrant the truth of the conclusion as in deductive reasoning, the conclusion nevertheless holds in most/typical/etc. cases in which the premises hold. illative tier: beyond truth-preservation typically: fixed minimal deductive (non-defeasible) standard of reasoning / core rules (e.g., classical logic) jump to more conclusions given additional warrants that allow for defeasible conclusions
dialectic tier: e.g., exceptional circumstances conflicting defeasible and deductive inferences/arguments
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
15 / 104
Ampliative approaches While the truth of the premises does not warrant the truth of the conclusion as in deductive reasoning, the conclusion nevertheless holds in most/typical/etc. cases in which the premises hold. illative tier: beyond truth-preservation typically: fixed minimal deductive (non-defeasible) standard of reasoning / core rules (e.g., classical logic) jump to more conclusions given additional warrants that allow for defeasible conclusions
dialectic tier: e.g., exceptional circumstances conflicting defeasible and deductive inferences/arguments
Examples inheritance networks default logic abductive logics inductive generalisation Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
15 / 104
Classical Logic only illative tier What’s the inference ticket in classical logic (in short, CL)? β follows from α iff in all classical interpretations in which α is true, also β is true entailment: ’truth-preservation’, ’deduction’ e.g. ∀x(P(x) → Q(x)) P(a) Q(a)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
16 / 104
Classical Logic only illative tier What’s the inference ticket in classical logic (in short, CL)? β follows from α iff in all classical interpretations in which α is true, also β is true entailment: ’truth-preservation’, ’deduction’ e.g. ∀x(P(x) → Q(x)) P(a) Q(a)
Hence . . . CL seems not apt to characterize reasoning that is dynamic in the sense that reasoners are prepared to retract inferences. Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
16 / 104
Summing up
corrective approach ampliative approach
Christian Straßer (RUB, UGENT)
illative tier deductive inferences ded. + non-ded. inf.
dialectic tier conflicting ded. inferencces confl. non-ded. inferences
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
17 / 104
Summing up
corrective approach ampliative approach
illative tier deductive inferences ded. + non-ded. inf.
dialectic tier conflicting ded. inferencces confl. non-ded. inferences
In practice: distinction is not so clear-cut
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
17 / 104
Pessimism in the 60ies
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
18 / 104
Toulmin Scheme (Toulmin (1958))
Premises
Backing
Christian Straßer (RUB, UGENT)
Conclusion
Warrant
Defeat
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
19 / 104
A Toulmin Argument
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
20 / 104
Nonmonotonic Logic to the Rescue
Figure: Artificial Intelligence, Volume 13, Issues 1–2, Pages 1-174,(April 1980), Special Issue on Non-Monotonic Logic
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
21 / 104
Nonmonotonic Logic to the Rescue
Figure: Artificial Intelligence, Volume 13, Issues 1–2, Pages 1-174,(April 1980), Special Issue on Non-Monotonic Logic
Aim capture defeasible reasoning in a mathematically precise way reproduce the success of CL in the domain of mathematical reasoning in the less sterile/idealized domain of defeasible reasoning where incompleteness and uncertainty play a central role Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
21 / 104
Nowadays
cooperations between formal and informal logicians e.g., Douglas Walton and formal argumentation (Gordon et al. (2007))
shift of normative standards in cognitive science, e.g. Stenning and Van Lambalgen (2008) Pfeifer (2014)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
22 / 104
Topic 1
2 3
4 5
Defeasible Reasoning – Some Basic Concepts Some examples to warm up Some conceptual distinctions Nonmonotonic Logic in Context The Dynamics of Defeasible Reasoning Default Logic (in the tradition of Reiter) Warming up Defaults and Default Theories Inferring Alternatives and more examples Meta-Properties Introducing Priorities Summing up Autoepistemic Logic Bibliography Bibliography
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
23 / 104
Nonmonotonicity Nonmonotonic Logics Monotony: If Γ |∼ φ then Γ ∪ Γ0 |∼ φ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
24 / 104
Nonmonotonicity Nonmonotonic Logics Monotony: If Γ |∼ φ then Γ ∪ Γ0 |∼ φ. If we define a consequence function Cn(•) by Cn(Γ) = {φ | Γ |∼ φ} we can equivalently express Monotony by: If φ ∈ Cn(Γ) then φ ∈ Cn(Γ ∪ Γ0 ).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
24 / 104
Nonmonotonicity Nonmonotonic Logics Monotony: If Γ |∼ φ then Γ ∪ Γ0 |∼ φ. If we define a consequence function Cn(•) by Cn(Γ) = {φ | Γ |∼ φ} we can equivalently express Monotony by: If φ ∈ Cn(Γ) then φ ∈ Cn(Γ ∪ Γ0 ).
External and Internal dynamics (Pollock / Batens) External Dynamics retract conclusions under the influence of new information
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
24 / 104
Nonmonotonicity Nonmonotonic Logics Monotony: If Γ |∼ φ then Γ ∪ Γ0 |∼ φ. If we define a consequence function Cn(•) by Cn(Γ) = {φ | Γ |∼ φ} we can equivalently express Monotony by: If φ ∈ Cn(Γ) then φ ∈ Cn(Γ ∪ Γ0 ).
External and Internal dynamics (Pollock / Batens) External Dynamics retract conclusions under the influence of new information
Internal Dynamics retract conclusions under the influence of a progressive analysis of the given information Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
24 / 104
What to replace Monotonicity with? Cautious Monotony If Γ |∼ ϕ and Γ |∼ ψ, then Γ, ϕ |∼ ψ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
25 / 104
What to replace Monotonicity with? Cautious Monotony If Γ |∼ ϕ and Γ |∼ ψ, then Γ, ϕ |∼ ψ. Cautious Monotony is the converse of Cut:
(Cautious) Cut If Γ |∼ ϕ and Γ, ϕ |∼ ψ then Γ |∼ ψ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
25 / 104
What to replace Monotonicity with? Cautious Monotony If Γ |∼ ϕ and Γ |∼ ψ, then Γ, ϕ |∼ ψ. Cautious Monotony is the converse of Cut:
(Cautious) Cut If Γ |∼ ϕ and Γ, ϕ |∼ ψ then Γ |∼ ψ. Cautious Monotonicity (Cut) states that adding a consequence ϕ back into the premise-set Γ does not lead to any decrease (increase) in inferential power.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
25 / 104
What to replace Monotonicity with? Cautious Monotony If Γ |∼ ϕ and Γ |∼ ψ, then Γ, ϕ |∼ ψ. Cautious Monotony is the converse of Cut:
(Cautious) Cut If Γ |∼ ϕ and Γ, ϕ |∼ ψ then Γ |∼ ψ. Cautious Monotonicity (Cut) states that adding a consequence ϕ back into the premise-set Γ does not lead to any decrease (increase) in inferential power. Both together tell us that inference is a cumulative enterprise: we can keep drawing consequences that can in turn be used as additional premises, without affecting the set of conclusions. Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
25 / 104
Core property
central place in non-monotonic logic Gabbay Gabbay (1985) Kraus, Lehmann, Magidor ("KLM") Kraus et al. (1990)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
26 / 104
Core property
central place in non-monotonic logic Gabbay Gabbay (1985) Kraus, Lehmann, Magidor ("KLM") Kraus et al. (1990)
also empirically confirmed property of actual reasoning
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
26 / 104
Core property
central place in non-monotonic logic Gabbay Gabbay (1985) Kraus, Lehmann, Magidor ("KLM") Kraus et al. (1990)
also empirically confirmed property of actual reasoning however: misses e.g., from Reiter’s Default Logic (see below)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
26 / 104
Be Cautious with Cut
(Cautious) Cut If Γ |∼ ϕ and Γ, ϕ |∼ ψ then Γ |∼ ψ.
. . . to be distinguished from: If Γ0 |∼ ϕ and Γ, ϕ |∼ ψ then Γ, Γ0 |∼ ψ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
27 / 104
Be Cautious with Cut
(Cautious) Cut If Γ |∼ ϕ and Γ, ϕ |∼ ψ then Γ |∼ ψ.
. . . to be distinguished from: If Γ0 |∼ ϕ and Γ, ϕ |∼ ψ then Γ, Γ0 |∼ ψ. Do you see why the second version of CUT is not suitable for nonmonotonic logic?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
27 / 104
What to replace Monotonicity with?
Rational Monotony If it’s not the case that Γ |∼ ¬ϕ, and moreover Γ |∼ ψ, then Γ, ϕ |∼ ψ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
28 / 104
What to replace Monotonicity with?
Rational Monotony If it’s not the case that Γ |∼ ¬ϕ, and moreover Γ |∼ ψ, then Γ, ϕ |∼ ψ. Adding formulas that are consistent with our current beliefs does not lead to a decrease in our set of conclusions.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
28 / 104
Stalnaker’s problematic example for Rational Monotonicity Rational Monotony If it’s not the case that Γ |∼ ¬ϕ, and moreover Γ |∼ ψ, then Γ, ϕ |∼ ψ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
29 / 104
Stalnaker’s problematic example for Rational Monotonicity Rational Monotony If it’s not the case that Γ |∼ ¬ϕ, and moreover Γ |∼ ψ, then Γ, ϕ |∼ ψ.
Stalnaker’s counter-example (Stalnaker (1994)) Stalnaker adopts the reading of ’α |∼ β ’ as ’Given our initial set of beliefs Γ, if we learn α then (nonmonotonically) infer β.’ proposed in Makinson and Gärdenfors (1991).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
29 / 104
Stalnaker’s problematic example for Rational Monotonicity Rational Monotony If it’s not the case that Γ |∼ ¬ϕ, and moreover Γ |∼ ψ, then Γ, ϕ |∼ ψ.
Stalnaker’s counter-example (Stalnaker (1994)) Stalnaker adopts the reading of ’α |∼ β ’ as ’Given our initial set of beliefs Γ, if we learn α then (nonmonotonically) infer β.’ proposed in Makinson and Gärdenfors (1991). Consider the three composers: Verdi (v), Bizet (b), and Satie (s), and suppose that we initially accept (correctly but defeasibly) that Verdi is Italian (I(v)), while Bizet and Satie are French (F(b), F(s)).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
29 / 104
Stalnaker’s problematic example for Rational Monotonicity Rational Monotony If it’s not the case that Γ |∼ ¬ϕ, and moreover Γ |∼ ψ, then Γ, ϕ |∼ ψ.
Stalnaker’s counter-example (Stalnaker (1994)) Stalnaker adopts the reading of ’α |∼ β ’ as ’Given our initial set of beliefs Γ, if we learn α then (nonmonotonically) infer β.’ proposed in Makinson and Gärdenfors (1991). Consider the three composers: Verdi (v), Bizet (b), and Satie (s), and suppose that we initially accept (correctly but defeasibly) that Verdi is Italian (I(v)), while Bizet and Satie are French (F(b), F(s)). Suppose now that we learn that Verdi and Bizet are compatriots (C(v,b)).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
29 / 104
Stalnaker’s problematic example for Rational Monotonicity Rational Monotony If it’s not the case that Γ |∼ ¬ϕ, and moreover Γ |∼ ψ, then Γ, ϕ |∼ ψ.
Stalnaker’s counter-example (Stalnaker (1994)) Stalnaker adopts the reading of ’α |∼ β ’ as ’Given our initial set of beliefs Γ, if we learn α then (nonmonotonically) infer β.’ proposed in Makinson and Gärdenfors (1991). Consider the three composers: Verdi (v), Bizet (b), and Satie (s), and suppose that we initially accept (correctly but defeasibly) that Verdi is Italian (I(v)), while Bizet and Satie are French (F(b), F(s)). Suppose now that we learn that Verdi and Bizet are compatriots (C(v,b)). then: C (v , b) |∼ F (s)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
29 / 104
Stalnaker’s problematic example for Rational Monotonicity Rational Monotony If it’s not the case that Γ |∼ ¬ϕ, and moreover Γ |∼ ψ, then Γ, ϕ |∼ ψ.
Stalnaker’s counter-example (Stalnaker (1994)) Stalnaker adopts the reading of ’α |∼ β ’ as ’Given our initial set of beliefs Γ, if we learn α then (nonmonotonically) infer β.’ proposed in Makinson and Gärdenfors (1991). Consider the three composers: Verdi (v), Bizet (b), and Satie (s), and suppose that we initially accept (correctly but defeasibly) that Verdi is Italian (I(v)), while Bizet and Satie are French (F(b), F(s)). Suppose now that we learn that Verdi and Bizet are compatriots (C(v,b)). then: C (v , b) |∼ F (s) Now consider: C(v,s), then C (v , b) 6|∼ ¬C (v , s). Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
29 / 104
Stalnaker’s problematic example for Rational Monotonicity Rational Monotony If it’s not the case that Γ |∼ ¬ϕ, and moreover Γ |∼ ψ, then Γ, ϕ |∼ ψ.
Stalnaker’s counter-example (Stalnaker (1994)) Stalnaker adopts the reading of ’α |∼ β ’ as ’Given our initial set of beliefs Γ, if we learn α then (nonmonotonically) infer β.’ proposed in Makinson and Gärdenfors (1991). Consider the three composers: Verdi (v), Bizet (b), and Satie (s), and suppose that we initially accept (correctly but defeasibly) that Verdi is Italian (I(v)), while Bizet and Satie are French (F(b), F(s)). Suppose now that we learn that Verdi and Bizet are compatriots (C(v,b)). then: C (v , b) |∼ F (s) Now consider: C(v,s), then C (v , b) 6|∼ ¬C (v , s). But if we add C(v,s) to our beliefs, then C (v , b), C (v , s) 6|∼ F (s). Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
29 / 104
Schematic Inference Graphs (e.g., Inheritance Nets, Formal Argumentation, etc.)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
30 / 104
Schematic Inference Graphs (e.g., Inheritance Nets, Formal Argumentation, etc.)
We use the following conventions: →: signify deductive or strict (i.e., non-defeasible) inferences, →: signify defeasible inferences, and
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
30 / 104
Schematic Inference Graphs (e.g., Inheritance Nets, Formal Argumentation, etc.)
We use the following conventions: →: signify deductive or strict (i.e., non-defeasible) inferences, →: signify defeasible inferences, and strikethrough (single resp. double) arrows signify that the negation of the pointed formula is (defeasibly resp. strictly) implied.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
30 / 104
Schematic Inference Graphs (e.g., Inheritance Nets, Formal Argumentation, etc.)
We use the following conventions: →: signify deductive or strict (i.e., non-defeasible) inferences, →: signify defeasible inferences, and strikethrough (single resp. double) arrows signify that the negation of the pointed formula is (defeasibly resp. strictly) implied. Arguments: e.g., Penguin → Bird → flies
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
30 / 104
Schematic Inference Graphs (e.g., Inheritance Nets, Formal Argumentation, etc.)
We use the following conventions: →: signify deductive or strict (i.e., non-defeasible) inferences, →: signify defeasible inferences, and strikethrough (single resp. double) arrows signify that the negation of the pointed formula is (defeasibly resp. strictly) implied. Arguments: e.g., Penguin → Bird → flies So, we can read the diagram as follows: Penguins are birds (no exceptions); Birds usually fly; and Penguins usually don’t fly. Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
30 / 104
Resolution: Strict beats defeasible
Two types of conflicts 1
2
conflicts between defeasible conclusions and "hard facts,": A → B vs. A0 → ¬B conflicts between one potential defeasible conclusion and another: A → B vs. A0 → ¬B
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
31 / 104
Resolution: Strict beats defeasible
Two types of conflicts 1
2
conflicts between defeasible conclusions and "hard facts,": A → B vs. A0 → ¬B conflicts between one potential defeasible conclusion and another: A → B vs. A0 → ¬B
Resolution Concerning 1: hard facts are prioritized. Concerning 2: this is more complicated . . .
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
31 / 104
Specificity / Preemption
According to the Specificity Principle an inference with a more specific antecedent overrides a conflicting defeasible inference with a less specific antecedent.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
32 / 104
Specificity / Preemption
According to the Specificity Principle an inference with a more specific antecedent overrides a conflicting defeasible inference with a less specific antecedent. lot of work in inheritance nets is devoted to this problem (see Horty (1994))
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
32 / 104
Topic 1
2 3
4 5
Defeasible Reasoning – Some Basic Concepts Some examples to warm up Some conceptual distinctions Nonmonotonic Logic in Context The Dynamics of Defeasible Reasoning Default Logic (in the tradition of Reiter) Warming up Defaults and Default Theories Inferring Alternatives and more examples Meta-Properties Introducing Priorities Summing up Autoepistemic Logic Bibliography Bibliography
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
33 / 104
Some References to Classical Articles
A logic for default reasoning. Artificial Intelligence, 1–2(13). Reiter (1980) A logical framework for default reasoning. Artificial intelligence, 36(1), 27–47. Poole (1988) The effect of knowledge on belief: conditioning, specificity and the lottery paradox in default reasoning. Artificial Intelligence, 49(1-3), 281–307. Poole (1991) Considerations on default logic: an alternative approach. Computational intelligence, 4(1), 1–16. Łukaszewicz (1988)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
34 / 104
Short Reminder: 1st order logic
Logical symbols quantifiers ∀, ∃ logical connectives ∧, ∨, ⊃, ¬ brackets variables
non-logical symbols predicate / relation symbols with specific arity function symbols with specific arity constants (0-ary functions)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
35 / 104
Short Reminder: 1st order logic, special terminology terms: variables, f (t1 , . . . , tn ) where ti are terms atomic formula: P(t1 , . . . , tn ) formulas: h∀, ∃, ∧, ∨, ⊃, ¬i closure of atomic formulas free / bound variables sentence: formula without free variables instance of a formula ϕ: substitution of some free variables for terms ground term: term without variables ground instance: instance that is a sentence (obtained by substituting all free variables by ground terms)
Example bird(Tweety) ⊃ flies(Tweety) is a ground instance of bird(x) ⊃ flies(x) Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
36 / 104
What’s a default conditional prerequisite α(x) : β1 (x), . . . , βn (b) γ(x)
where x = x1 , . . . , xm , and α(x), β1 (x), . . . , βn (x), γ(x) are formulas whose free variables are among x1 , . . . , xm .
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
37 / 104
What’s a default conditional prerequisite
justification
α(x) : β1 (x), . . . , βn (b) γ(x)
where x = x1 , . . . , xm , and α(x), β1 (x), . . . , βn (x), γ(x) are formulas whose free variables are among x1 , . . . , xm .
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
37 / 104
What’s a default conditional prerequisite
justification
α(x) : β1 (x), . . . , βn (b) γ(x) conclusion where x = x1 , . . . , xm , and α(x), β1 (x), . . . , βn (x), γ(x) are formulas whose free variables are among x1 , . . . , xm .
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
37 / 104
What’s a default conditional prerequisite
justification
α(x) : β1 (x), . . . , βn (b) γ(x) conclusion where x = x1 , . . . , xm , and α(x), β1 (x), . . . , βn (x), γ(x) are formulas whose free variables are among x1 , . . . , xm .
Application of a default The default is applied in order to derive the c-ground instance of γ in case trigger: α(c) belongs to our set of —depending on the perspective we have— beliefs/(defeasible) knowledge/plausible assumptions/etc. (henceforth I will speak only about beliefs) justification: the set of our beliefs is consistent with each βi (c) Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
37 / 104
Default Theory
h∆, Φi set of defaults
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
38 / 104
Default Theory
set of ’facts’ h∆, Φi set of defaults
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
38 / 104
Default Theory
set of ’facts’ h∆, Φi set of defaults
simple example
bird(x) : flies(x) flies(x) Φ = {bird(Tweety), cat(Sylvester)} ∆=
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
38 / 104
Types of defaults Normal defaults α(x) : γ(x) γ(x) rather natural representation
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
39 / 104
Types of defaults Normal defaults α(x) : γ(x) γ(x) rather natural representation
Semi-Normal defaults α(x) : β(x) γ(x) where β(x) ` γ(x). E.g., α(x) : γ(x) ∧ β(x) γ(x) Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
39 / 104
How to reason with default theories?
Idea Apply iteratively modus ponens to defaults. This way build step-wise an extension (sets of beliefs that are obtained in this way)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
40 / 104
Here’s how it goes: guess the extension Ξ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
41 / 104
Here’s how it goes: guess the extension Ξ init beliefs: Ξ? = Φ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
41 / 104
Here’s how it goes: guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether:
Christian Straßer (RUB, UGENT)
α(x) : β(x) γ(x)
Tutorial: Nonmonotonic Logic (Day 1)
∈ ∆ and
September 2, 2015
41 / 104
Here’s how it goes: guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
41 / 104
Here’s how it goes: guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
41 / 104
Here’s how it goes: guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
41 / 104
Here’s how it goes: guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)} if no:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
41 / 104
Here’s how it goes: guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)} if no: try another triggered default in ∆ (goto (†))
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
41 / 104
Here’s how it goes: guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)} if no: try another triggered default in ∆ (goto (†)) if there isn’t: terminate.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
41 / 104
Here’s how it goes: guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)} if no: try another triggered default in ∆ (goto (†)) if there isn’t: terminate. if Ξ = Cn(Ξ? ): extension found.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
41 / 104
Here’s how it goes: guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)} if no: try another triggered default in ∆ (goto (†)) if there isn’t: terminate. if Ξ = Cn(Ξ? ): extension found.
Problem (?) We have to guess and use our guess when adding new defaults. Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
41 / 104
Example: Tweety Let T = h∆, Φi where bird(x) : flies(x) ∆= flies(x)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
42 / 104
Example: Tweety Let T = h∆, Φi where bird(x) : flies(x) ∆= flies(x) and Φ = {bird(Tweety), cat(Sylvester)}.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
42 / 104
Example: Tweety Let T = h∆, Φi where bird(x) : flies(x) ∆= flies(x) and Φ = {bird(Tweety), cat(Sylvester)}. we have only the two constants Tweety and Sylvester in the language
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
42 / 104
Example: Tweety Let T = h∆, Φi where bird(x) : flies(x) ∆= flies(x) and Φ = {bird(Tweety), cat(Sylvester)}. we have only the two constants Tweety and Sylvester in the language
Building up the extensions: guess: Ξ = Cn({flies(Tweety)} ∪ Φ})
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
42 / 104
Example: Tweety Let T = h∆, Φi where bird(x) : flies(x) ∆= flies(x) and Φ = {bird(Tweety), cat(Sylvester)}. we have only the two constants Tweety and Sylvester in the language
Building up the extensions: guess: Ξ = Cn({flies(Tweety)} ∪ Φ}) our initial knowledge is Φ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
42 / 104
Example: Tweety Let T = h∆, Φi where bird(x) : flies(x) ∆= flies(x) and Φ = {bird(Tweety), cat(Sylvester)}. we have only the two constants Tweety and Sylvester in the language
Building up the extensions: guess: Ξ = Cn({flies(Tweety)} ∪ Φ}) our initial knowledge is Φ note that the Sylvester-instance of our default is not applicable to Φ since Φ 0 bird(Sylvester)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
42 / 104
Example: Tweety Let T = h∆, Φi where bird(x) : flies(x) ∆= flies(x) and Φ = {bird(Tweety), cat(Sylvester)}. we have only the two constants Tweety and Sylvester in the language
Building up the extensions: guess: Ξ = Cn({flies(Tweety)} ∪ Φ}) our initial knowledge is Φ note that the Sylvester-instance of our default is not applicable to Φ since Φ 0 bird(Sylvester) however, we have bird(Tweety) and fly(Tweety) is consistent with Ξ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
42 / 104
Example: Tweety Let T = h∆, Φi where bird(x) : flies(x) ∆= flies(x) and Φ = {bird(Tweety), cat(Sylvester)}. we have only the two constants Tweety and Sylvester in the language
Building up the extensions: guess: Ξ = Cn({flies(Tweety)} ∪ Φ}) our initial knowledge is Φ note that the Sylvester-instance of our default is not applicable to Φ since Φ 0 bird(Sylvester) however, we have bird(Tweety) and fly(Tweety) is consistent with Ξ. fixed point reached
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
42 / 104
Example: Tweety Let T = h∆, Φi where bird(x) : flies(x) ∆= flies(x) and Φ = {bird(Tweety), cat(Sylvester)}. we have only the two constants Tweety and Sylvester in the language
Building up the extensions: guess: Ξ = Cn({flies(Tweety)} ∪ Φ}) our initial knowledge is Φ note that the Sylvester-instance of our default is not applicable to Φ since Φ 0 bird(Sylvester) however, we have bird(Tweety) and fly(Tweety) is consistent with Ξ. fixed point reached the only extension is Ξ. Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
42 / 104
Question
Are extensions always unique?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
43 / 104
The Nixon Diamond
Let T = h∆, Φi where ∆= n
quaker(x) : pacifist(x) republican(x) : ¬pacifist(x) , pacifist(x) ¬pacifist(x)
o
Φ = {Quaker(Nixon), republican(Nixon)}.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
44 / 104
The Nixon Diamond
Let T = h∆, Φi where ∆= n
quaker(x) : pacifist(x) republican(x) : ¬pacifist(x) , pacifist(x) ¬pacifist(x)
o
Φ = {Quaker(Nixon), republican(Nixon)}.
There are two extensions:
Christian Straßer (RUB, UGENT)
1
one that contains pacifist(Nixon),
2
and one that contains ¬pacifist(Nixon).
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
44 / 104
Another example Let h∆, Φi be a default theory where α(x) : β(x) γ(x) : ¬β(x) , ∆= γ(x) ¬β(x) Φ = {α(c)}
Question Is there an extension?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
45 / 104
Another example Let h∆, Φi be a default theory where α(x) : β(x) γ(x) : ¬β(x) , ∆= γ(x) ¬β(x) Φ = {α(c)}
Question Is there an extension?
Guess: Cn({α(c), γ(c)})
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
45 / 104
Another example Let h∆, Φi be a default theory where α(x) : β(x) γ(x) : ¬β(x) , ∆= γ(x) ¬β(x) Φ = {α(c)}
Question Is there an extension?
Guess: Cn({α(c), γ(c)}) first run: Φ? = Φ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
45 / 104
Another example Let h∆, Φi be a default theory where α(x) : β(x) γ(x) : ¬β(x) , ∆= γ(x) ¬β(x) Φ = {α(c)}
Question Is there an extension?
Guess: Cn({α(c), γ(c)}) first run: Φ? = Φ α(x) : β(x) γ(x)
is triggered and justified: apply
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
45 / 104
Another example Let h∆, Φi be a default theory where α(x) : β(x) γ(x) : ¬β(x) , ∆= γ(x) ¬β(x) Φ = {α(c)}
Question Is there an extension?
Guess: Cn({α(c), γ(c)}) first run: Φ? = Φ α(x) : β(x) γ(x)
is triggered and justified: apply Φ? = Φ ∪ {γ(c)} Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
45 / 104
Another example Let h∆, Φi be a default theory where α(x) : β(x) γ(x) : ¬β(x) , ∆= γ(x) ¬β(x) Φ = {α(c)}
Question Is there an extension?
Guess: Cn({α(c), γ(c)}) second run:
first run:
γ(x) : ¬β(x) ¬β(x)
Φ? = Φ α(x) : β(x) γ(x)
is triggered and justified: apply
is triggered and
justified
Φ? = Φ ∪ {γ(c)} Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
45 / 104
Another example Let h∆, Φi be a default theory where α(x) : β(x) γ(x) : ¬β(x) , ∆= γ(x) ¬β(x) Φ = {α(c)}
Question Is there an extension?
Guess: Cn({α(c), γ(c)}) second run:
first run:
γ(x) : ¬β(x) ¬β(x)
Φ? = Φ α(x) : β(x) γ(x)
is triggered and justified: apply
is triggered and
justified
Φ? = Φ ∪ {γ(c), ¬β(c)}
Φ? = Φ ∪ {γ(c)} Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
45 / 104
Another example Let h∆, Φi be a default theory where α(x) : β(x) γ(x) : ¬β(x) , ∆= γ(x) ¬β(x) Φ = {α(c)}
Question Is there an extension?
Guess: Cn({α(c), γ(c)}) second run:
first run:
γ(x) : ¬β(x) ¬β(x)
Φ? = Φ α(x) : β(x) γ(x)
is triggered and justified: apply Φ? = Φ ∪ {γ(c)} Christian Straßer (RUB, UGENT)
is triggered and
justified
Φ? = Φ ∪ {γ(c), ¬β(c)} our guess is wrong (similar problems with other guesses)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
45 / 104
How to define the consequences of a default theory?
Two approaches:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
46 / 104
How to define the consequences of a default theory?
Two approaches:
Skeptical approach h∆, Φi `skp A iff A ∈
Christian Straßer (RUB, UGENT)
T
Extensions(h∆, Φi)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
46 / 104
How to define the consequences of a default theory?
Two approaches:
Skeptical approach h∆, Φi `skp A iff A ∈
T
Extensions(h∆, Φi)
S
Extensions(h∆, Φi)
Credulous approach h∆, Φi `crd A iff A ∈
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
46 / 104
How to define the consequences of a default theory?
Two approaches:
Skeptical approach h∆, Φi `skp A iff A ∈
T
Extensions(h∆, Φi)
S
Extensions(h∆, Φi)
Credulous approach h∆, Φi `crd A iff A ∈
Question: When is which approach useful?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
46 / 104
Alternative: Consistency Check in the end guess the extension Ξ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
47 / 104
Alternative: Consistency Check in the end guess the extension Ξ init beliefs: Ξ? = Φ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
47 / 104
Alternative: Consistency Check in the end guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether:
Christian Straßer (RUB, UGENT)
α(x) : β(x) γ(x)
Tutorial: Nonmonotonic Logic (Day 1)
∈ ∆ and
September 2, 2015
47 / 104
Alternative: Consistency Check in the end guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
47 / 104
Alternative: Consistency Check in the end guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
47 / 104
Alternative: Consistency Check in the end guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
47 / 104
Alternative: Consistency Check in the end guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)} if no:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
47 / 104
Alternative: Consistency Check in the end guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)} if no:
try another triggered default in ∆ (goto (†))
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
47 / 104
Alternative: Consistency Check in the end guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)} if no:
try another triggered default in ∆ (goto (†)) if there isn’t: terminate
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
47 / 104
Alternative: Consistency Check in the end guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)} if no:
try another triggered default in ∆ (goto (†)) if there isn’t: terminate
if Ξ? is consistent with all the justifications of the applied defaults and if Ξ = Cn(Ξ? ) we found an extension, otherwise try again from the start.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
47 / 104
Alternative: Consistency Check in the end guess the extension Ξ init beliefs: Ξ? = Φ (†) take an c-ground instance of an default check whether: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
trigger?: Ξ? ` α(c) conflicted?: each βi (c) (1 ≤ i ≤ n) is consistent with Ξ (!!)
if yes: update beliefs: Ξ? := Ξ? ∪ {γ(c)} if no:
try another triggered default in ∆ (goto (†)) if there isn’t: terminate
if Ξ? is consistent with all the justifications of the applied defaults and if Ξ = Cn(Ξ? ) we found an extension, otherwise try again from the start. no initial guess guessing on-the-fly when we choose which triggered defaults to apply only in the end we see whether we guessed luckily Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
47 / 104
Going strictly procedural with Łukaszewicz (1988) Let h∆, Φi be a default theory. init: Φ? = Φ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
48 / 104
Going strictly procedural with Łukaszewicz (1988) Let h∆, Φi be a default theory. init: Φ? = Φ (†) take a c-instance of an arbitrary default check:
Christian Straßer (RUB, UGENT)
α(x) : β(x) γ(x)
Tutorial: Nonmonotonic Logic (Day 1)
∈ ∆ and
September 2, 2015
48 / 104
Going strictly procedural with Łukaszewicz (1988) Let h∆, Φi be a default theory. init: Φ? = Φ (†) take a c-instance of an arbitrary default check: 1
α(x) : β(x) γ(x)
∈ ∆ and
Φ? ` α(c) (trigger)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
48 / 104
Going strictly procedural with Łukaszewicz (1988) Let h∆, Φi be a default theory. init: Φ? = Φ (†) take a c-instance of an arbitrary default check: 1 2
α(x) : β(x) γ(x)
∈ ∆ and
Φ? ` α(c) (trigger) each βi (c) is consistent with Φ? (justification 1)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
48 / 104
Going strictly procedural with Łukaszewicz (1988) Let h∆, Φi be a default theory. init: Φ? = Φ (†) take a c-instance of an arbitrary default check: 1 2 3
α(x) : β(x) γ(x)
∈ ∆ and
Φ? ` α(c) (trigger) each βi (c) is consistent with Φ? (justification 1) each justification of previously applied defaults is consistent with Φ? ∪ {γ(c)} (justification 2)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
48 / 104
Going strictly procedural with Łukaszewicz (1988) Let h∆, Φi be a default theory. init: Φ? = Φ (†) take a c-instance of an arbitrary default check: 1 2 3
α(x) : β(x) γ(x)
∈ ∆ and
Φ? ` α(c) (trigger) each βi (c) is consistent with Φ? (justification 1) each justification of previously applied defaults is consistent with Φ? ∪ {γ(c)} (justification 2)
if yes: Φ? = Φ? ∪ {γ(c)} and goto (†)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
48 / 104
Going strictly procedural with Łukaszewicz (1988) Let h∆, Φi be a default theory. init: Φ? = Φ (†) take a c-instance of an arbitrary default check: 1 2 3
α(x) : β(x) γ(x)
∈ ∆ and
Φ? ` α(c) (trigger) each βi (c) is consistent with Φ? (justification 1) each justification of previously applied defaults is consistent with Φ? ∪ {γ(c)} (justification 2)
if yes: Φ? = Φ? ∪ {γ(c)} and goto (†) if no:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
48 / 104
Going strictly procedural with Łukaszewicz (1988) Let h∆, Φi be a default theory. init: Φ? = Φ (†) take a c-instance of an arbitrary default check: 1 2 3
α(x) : β(x) γ(x)
∈ ∆ and
Φ? ` α(c) (trigger) each βi (c) is consistent with Φ? (justification 1) each justification of previously applied defaults is consistent with Φ? ∪ {γ(c)} (justification 2)
if yes: Φ? = Φ? ∪ {γ(c)} and goto (†) if no: if there is another instance of a default in ∆ that wasn’t tasted, goto (†) and test it
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
48 / 104
Going strictly procedural with Łukaszewicz (1988) Let h∆, Φi be a default theory. init: Φ? = Φ (†) take a c-instance of an arbitrary default check: 1 2 3
α(x) : β(x) γ(x)
∈ ∆ and
Φ? ` α(c) (trigger) each βi (c) is consistent with Φ? (justification 1) each justification of previously applied defaults is consistent with Φ? ∪ {γ(c)} (justification 2)
if yes: Φ? = Φ? ∪ {γ(c)} and goto (†) if no: if there is another instance of a default in ∆ that wasn’t tasted, goto (†) and test it otherwise: if Ξ = Cn(Φ? ), we found an extension.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
48 / 104
Some properties of the new procedure No guess needed. real procedural character guarantees existence of an extension hence: yields sometimes different results from Reiter’s account
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
49 / 104
Some properties of the new procedure No guess needed. real procedural character guarantees existence of an extension hence: yields sometimes different results from Reiter’s account
Question What happens in the new approach when plugging in the default theory h∆, Φi where α(x) : β(x) ∧ γ(x) γ(x) : ¬β(x) ∆= , γ(x) ¬β(x) Φ = {α(c)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
49 / 104
Recall the three conditions 1 2 3
Φ? ` α(c) (trigger) ?
each βi (c) is consistent with Φ (justification 1) each justification of previously applied defaults is consistent with Φ? ∪ {γ(c)} (justification 2)
∆= n
α(x) : β(x)∧γ(x) γ(x) : ¬β(x) , γ(x) ¬β(x)
Φ = {α(c)}
o
Non-Procedural Fixed-Point Characterisations What about the following definition?
Definition: Extension’ Ξ is an extension’ of a default theory h∆, Φi iff it is a minimal set that satisfies the following conditions:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
50 / 104
Non-Procedural Fixed-Point Characterisations What about the following definition?
Definition: Extension’ Ξ is an extension’ of a default theory h∆, Φi iff it is a minimal set that satisfies the following conditions: 1
Φ⊆Ξ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
50 / 104
Non-Procedural Fixed-Point Characterisations What about the following definition?
Definition: Extension’ Ξ is an extension’ of a default theory h∆, Φi iff it is a minimal set that satisfies the following conditions: 1
Φ⊆Ξ
2
Cn(Ξ) = Ξ (fixed-point)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
50 / 104
Non-Procedural Fixed-Point Characterisations What about the following definition?
Definition: Extension’ Ξ is an extension’ of a default theory h∆, Φi iff it is a minimal set that satisfies the following conditions: 1
Φ⊆Ξ
2
Cn(Ξ) = Ξ (fixed-point)
3
if
α(c) : β(c) γ(c)
is a c-instance of some default in ∆ and
then γ(c) ∈ Ξ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
50 / 104
Non-Procedural Fixed-Point Characterisations What about the following definition?
Definition: Extension’ Ξ is an extension’ of a default theory h∆, Φi iff it is a minimal set that satisfies the following conditions: 1
Φ⊆Ξ
2
Cn(Ξ) = Ξ (fixed-point)
3
if
α(c) : β(c) γ(c) 1
is a c-instance of some default in ∆ and
α(c) ∈ Ξ (trigger)
then γ(c) ∈ Ξ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
50 / 104
Non-Procedural Fixed-Point Characterisations What about the following definition?
Definition: Extension’ Ξ is an extension’ of a default theory h∆, Φi iff it is a minimal set that satisfies the following conditions: 1
Φ⊆Ξ
2
Cn(Ξ) = Ξ (fixed-point)
3
if
α(c) : β(c) γ(c) 1 2
is a c-instance of some default in ∆ and
α(c) ∈ Ξ (trigger) βi (c) is consistent with Ξ for all 1 ≤ i ≤ n (justification)
then γ(c) ∈ Ξ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
50 / 104
Non-Procedural Fixed-Point Characterisations What about the following definition?
Definition: Extension’ Ξ is an extension’ of a default theory h∆, Φi iff it is a minimal set that satisfies the following conditions: 1
Φ⊆Ξ
2
Cn(Ξ) = Ξ (fixed-point)
3
if
α(c) : β(c) γ(c) 1 2
is a c-instance of some default in ∆ and
α(c) ∈ Ξ (trigger) βi (c) is consistent with Ξ for all 1 ≤ i ≤ n (justification)
then γ(c) ∈ Ξ
Question Is this equivalent to the procedural approach? Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
50 / 104
Difference: Grounding
n o , ∅}i. Take h >:p p Note that Cn({¬p}) is a minimal set satisfying the previous conditions.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
51 / 104
Difference: Grounding
n o , ∅}i. Take h >:p p Note that Cn({¬p}) is a minimal set satisfying the previous conditions. However, the only extension is Cn({p}).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
51 / 104
Difference: Grounding
n o , ∅}i. Take h >:p p Note that Cn({¬p}) is a minimal set satisfying the previous conditions. However, the only extension is Cn({p}). We face the
Problem of grounding We expect that all members of the extension can be generated iteratively by chaining and detaching defaults.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
51 / 104
Non-Procedural Fixed-Point Characterisations
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
52 / 104
Non-Procedural Fixed-Point Characterisations
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying: 1
Φ ⊆ πΦ (Γ)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
52 / 104
Non-Procedural Fixed-Point Characterisations
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying: 1
Φ ⊆ πΦ (Γ)
2
πΦ (Γ) = Cn(πΦ (Γ)) (fixed point)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
52 / 104
Non-Procedural Fixed-Point Characterisations
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying: 1
Φ ⊆ πΦ (Γ)
2
πΦ (Γ) = Cn(πΦ (Γ)) (fixed point)
3
if
α(c) : β(c) γ(c)
is a c-instance of some default in ∆ and
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
52 / 104
Non-Procedural Fixed-Point Characterisations
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying: 1
Φ ⊆ πΦ (Γ)
2
πΦ (Γ) = Cn(πΦ (Γ)) (fixed point)
3
if
α(c) : β(c) γ(c) 1
is a c-instance of some default in ∆ and
α(c) ∈ πΦ (Γ) (trigger)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
52 / 104
Non-Procedural Fixed-Point Characterisations
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying: 1
Φ ⊆ πΦ (Γ)
2
πΦ (Γ) = Cn(πΦ (Γ)) (fixed point)
3
if
α(c) : β(c) γ(c) 1 2
is a c-instance of some default in ∆ and
α(c) ∈ πΦ (Γ) (trigger) ¬βi (c) ∈ / Γ for all 1 ≤ i ≤ n then γ(c) ∈ πΦ (Γ) (justification).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
52 / 104
Non-Procedural Fixed-Point Characterisations
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying: 1
Φ ⊆ πΦ (Γ)
2
πΦ (Γ) = Cn(πΦ (Γ)) (fixed point)
3
if
α(c) : β(c) γ(c) 1 2
is a c-instance of some default in ∆ and
α(c) ∈ πΦ (Γ) (trigger) ¬βi (c) ∈ / Γ for all 1 ≤ i ≤ n then γ(c) ∈ πΦ (Γ) (justification).
Definition: Extension A set of formulas Γ is an extension of h∆, Φi iff π(Γ) = Γ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
52 / 104
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
53 / 104
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying: 1
Φ ⊆ πΦ (Γ)
2
πΦ (Γ) = Cn(πΦ (Γ)) (fixed point)
3
if
α(c) : β(c) γ(c) 1 2
is a c-instance of some default in ∆ and
α(c) ∈ πΦ (Γ) (trigger) ¬βi (c) ∈ / Γ for all 1 ≤ i ≤ n then γ(c) ∈ πΦ (Γ) (justification).
Proof of the existence of π(Γ).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
53 / 104
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying: 1
Φ ⊆ πΦ (Γ)
2
πΦ (Γ) = Cn(πΦ (Γ)) (fixed point)
3
if
α(c) : β(c) γ(c) 1 2
is a c-instance of some default in ∆ and
α(c) ∈ πΦ (Γ) (trigger) ¬βi (c) ∈ / Γ for all 1 ≤ i ≤ n then γ(c) ∈ πΦ (Γ) (justification).
Proof of the existence of π(Γ). Let S be all sets that satisfy (1)–(3). (Note S 6= ∅ since L ∈ S.)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
53 / 104
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying: 1
Φ ⊆ πΦ (Γ)
2
πΦ (Γ) = Cn(πΦ (Γ)) (fixed point)
3
if
α(c) : β(c) γ(c) 1 2
is a c-instance of some default in ∆ and
α(c) ∈ πΦ (Γ) (trigger) ¬βi (c) ∈ / Γ for all 1 ≤ i ≤ n then γ(c) ∈ πΦ (Γ) (justification).
Proof of the existence of π(Γ). Let S be T all sets that satisfy (1)–(3). (Note S 6= ∅ since L ∈ S.) Let Γ0 = S. We have to show (1)–(3). 1
trivial
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
53 / 104
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying: 1
Φ ⊆ πΦ (Γ)
2
πΦ (Γ) = Cn(πΦ (Γ)) (fixed point)
3
if
α(c) : β(c) γ(c) 1 2
is a c-instance of some default in ∆ and
α(c) ∈ πΦ (Γ) (trigger) ¬βi (c) ∈ / Γ for all 1 ≤ i ≤ n then γ(c) ∈ πΦ (Γ) (justification).
Proof of the existence of π(Γ). Let S be T all sets that satisfy (1)–(3). (Note S 6= ∅ since L ∈ S.) Let Γ0 = S. We have to show (1)–(3). 1 2
trivial Suppose A ∈ Cn(Γ0 ). Hence (by monotonicity), Γ00 ` A for all Γ00 ∈ S. T Since Cn(Γ00 ) = Γ00 , A ∈ Γ00 . Thus, A ∈ S.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
53 / 104
Let h∆, Φi be a default theory. Define the operator πΦ such that for any set of formulas Γ, πΦ (Γ) the smallest set satisfying: 1
Φ ⊆ πΦ (Γ)
2
πΦ (Γ) = Cn(πΦ (Γ)) (fixed point)
3
if
α(c) : β(c) γ(c) 1 2
is a c-instance of some default in ∆ and
α(c) ∈ πΦ (Γ) (trigger) ¬βi (c) ∈ / Γ for all 1 ≤ i ≤ n then γ(c) ∈ πΦ (Γ) (justification).
Proof of the existence of π(Γ). Let S be T all sets that satisfy (1)–(3). (Note S 6= ∅ since L ∈ S.) Let Γ0 = S. We have to show (1)–(3). 1 2
3
trivial Suppose A ∈ Cn(Γ0 ). Hence (by monotonicity), Γ00 ` A for all Γ00 ∈ S. T Since Cn(Γ00 ) = Γ00 , A ∈ Γ00 . Thus, A ∈ S. Suppose α(c) ∈ Γ0 and ¬βi (c) ∈ / Γ for all i ≤ n. Hence, α(c) ∈ Γ00 for 00 00 all Γ ∈ S and thus γ(c) ∈ Γ . Thus, γ(c) ∈ Γ0 .
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
53 / 104
Cautious Cut If h∆, Φi ` A and h∆, Φ ∪ {A}i ` B then h∆, Φi ` B.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
54 / 104
Cautious Cut If h∆, Φi ` A and h∆, Φ ∪ {A}i ` B then h∆, Φi ` B. What do you think?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
54 / 104
Cautious Cut If h∆, Φi ` A and h∆, Φ ∪ {A}i ` B then h∆, Φi ` B. What do you think?
Lemma (from this Cut follows immediately for skeptical consequence) If h∆, Φi ` A, Ext(h∆, Φi) ⊆ Ext(h∆, Φ ∪ {A}i).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
54 / 104
Cautious Cut If h∆, Φi ` A and h∆, Φ ∪ {A}i ` B then h∆, Φi ` B. What do you think?
Lemma (from this Cut follows immediately for skeptical consequence) If h∆, Φi ` A, Ext(h∆, Φi) ⊆ Ext(h∆, Φ ∪ {A}i).
Proof Suppose h∆, Φi ` A and let Ξ ∈ Ext(h∆, Φi).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
54 / 104
Cautious Cut If h∆, Φi ` A and h∆, Φ ∪ {A}i ` B then h∆, Φi ` B. What do you think?
Lemma (from this Cut follows immediately for skeptical consequence) If h∆, Φi ` A, Ext(h∆, Φi) ⊆ Ext(h∆, Φ ∪ {A}i).
Proof Suppose h∆, Φi ` A and let Ξ ∈ Ext(h∆, Φi). We know that Ξ = πΦ (Ξ) and have to show that Ξ = πΦ∪{A} (Ξ).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
54 / 104
Cautious Cut If h∆, Φi ` A and h∆, Φ ∪ {A}i ` B then h∆, Φi ` B. What do you think?
Lemma (from this Cut follows immediately for skeptical consequence) If h∆, Φi ` A, Ext(h∆, Φi) ⊆ Ext(h∆, Φ ∪ {A}i).
Proof Suppose h∆, Φi ` A and let Ξ ∈ Ext(h∆, Φi). We know that Ξ = πΦ (Ξ) and have to show that Ξ = πΦ∪{A} (Ξ). Clearly, since A ∈ Ξ, Ξ satisfies (1)–(3) (relative to Φ ∪ {A}).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
54 / 104
Cautious Cut If h∆, Φi ` A and h∆, Φ ∪ {A}i ` B then h∆, Φi ` B. What do you think?
Lemma (from this Cut follows immediately for skeptical consequence) If h∆, Φi ` A, Ext(h∆, Φi) ⊆ Ext(h∆, Φ ∪ {A}i).
Proof Suppose h∆, Φi ` A and let Ξ ∈ Ext(h∆, Φi). We know that Ξ = πΦ (Ξ) and have to show that Ξ = πΦ∪{A} (Ξ). Clearly, since A ∈ Ξ, Ξ satisfies (1)–(3) (relative to Φ ∪ {A}). Assume πΦ∪{A} (Ξ) ⊂ Ξ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
54 / 104
Cautious Cut If h∆, Φi ` A and h∆, Φ ∪ {A}i ` B then h∆, Φi ` B. What do you think?
Lemma (from this Cut follows immediately for skeptical consequence) If h∆, Φi ` A, Ext(h∆, Φi) ⊆ Ext(h∆, Φ ∪ {A}i).
Proof Suppose h∆, Φi ` A and let Ξ ∈ Ext(h∆, Φi). We know that Ξ = πΦ (Ξ) and have to show that Ξ = πΦ∪{A} (Ξ). Clearly, since A ∈ Ξ, Ξ satisfies (1)–(3) (relative to Φ ∪ {A}). Assume πΦ∪{A} (Ξ) ⊂ Ξ. But then πΦ∪{A} (Ξ) also satisfies (1)–(3) relative to Φ which contradicts Ξ = πΦ (Ξ). Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
54 / 104
Cautious Cut If h∆, Φi ` A and h∆, Φ ∪ {A}i ` B then h∆, Φi ` B. What do you think?
Lemma (from this Cut follows immediately for skeptical consequence) If h∆, Φi ` A, Ext(h∆, Φi) ⊆ Ext(h∆, Φ ∪ {A}i).
Proof Suppose h∆, Φi ` A and let Ξ ∈ Ext(h∆, Φi). We know that Ξ = πΦ (Ξ) and have to show that Ξ = πΦ∪{A} (Ξ). Clearly, since A ∈ Ξ, Ξ satisfies (1)–(3) (relative to Φ ∪ {A}). Assume πΦ∪{A} (Ξ) ⊂ Ξ. But then πΦ∪{A} (Ξ) also satisfies (1)–(3) relative to Φ which contradicts Ξ = πΦ (Ξ). Hence, Ξ = πΦ∪{A} (Ξ). Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
54 / 104
Cautious Cut for credulous version?
Recall: Lemma If h∆, Φi ` A, Ext(h∆, Φi) ⊆ Ext(h∆, Φ ∪ {A}i).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
55 / 104
Cautious Cut for credulous version?
Recall: Lemma If h∆, Φi ` A, Ext(h∆, Φi) ⊆ Ext(h∆, Φ ∪ {A}i). What do you think, does this help?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
55 / 104
Cautious Cut for credulous version?
Recall: Lemma If h∆, Φi ` A, Ext(h∆, Φi) ⊆ Ext(h∆, Φ ∪ {A}i). What do you think, does this help?
Counter-example Take h∆, Φi where ∆ =
n
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
>:p p∨q:¬p p , ¬p
o .
September 2, 2015
55 / 104
Cautious Cut for credulous version?
Recall: Lemma If h∆, Φi ` A, Ext(h∆, Φi) ⊆ Ext(h∆, Φ ∪ {A}i). What do you think, does this help?
Counter-example Take h∆, Φi where ∆ =
n
>:p p∨q:¬p p , ¬p
o .
h∆, Φi `cred p ∨ q. h∆, Φ ∪ {p ∨ q}i `cred ¬p. But, h∆, Φi 0cred ¬p.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
55 / 104
The special status of normal default theories
A normal default theory is a default theory that only consists of normal defaults.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
56 / 104
The special status of normal default theories
A normal default theory is a default theory that only consists of normal defaults. A normal default theory always has an extension both in Reiter’s and in Lukaszewicz’s approach.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
56 / 104
The special status of normal default theories
A normal default theory is a default theory that only consists of normal defaults. A normal default theory always has an extension both in Reiter’s and in Lukaszewicz’s approach. For normal theories the set of Reiter extensions and the set of Lukaszewicz extensions coincides.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
56 / 104
Floating conclusions
Task 1
2
Christian Straßer (RUB, UGENT)
What are the two extensions of this default theory? Is politically−motivated(Nixon) derivable?
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
57 / 104
{ Nixon, quaker, republican, dove, hawk, politically motivated }
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
58 / 104
{ Nixon, quaker, republican, dove, hawk, politically motivated }
Christian Straßer (RUB, UGENT)
{ Nixon, quaker, republican, dove, ¬hawk, politically motivated }
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
58 / 104
{ Nixon, quaker, republican, dove, hawk, politically motivated }
Christian Straßer (RUB, UGENT)
{ Nixon, quaker, republican, dove, ¬hawk, politically motivated }
Tutorial: Nonmonotonic Logic (Day 1)
{ Nixon, quaker, republican, ¬dove, hawk, politically motivated }
September 2, 2015
58 / 104
Specificity
Question Is flies(Tweety) derivable?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
59 / 104
Specificity
Question Is flies(Tweety) derivable?
Nope There are two extensions:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
59 / 104
Specificity
Question Is flies(Tweety) derivable?
Nope There are two extensions: 1
one with flies(Tweety)
2
one with ¬flies(Tweety)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
59 / 104
Lukaszewicz’s Fishing example Let T = h∆, Φi where ∆= n
Sunday : I−go−fishing∧¬I−wake−up−late Holidays : I−wake−up−late , I−go−fishing I−wake−up−late
o
Φ = {Sunday, Holidays}.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
60 / 104
Lukaszewicz’s Fishing example Let T = h∆, Φi where ∆= n
Sunday : I−go−fishing∧¬I−wake−up−late Holidays : I−wake−up−late , I−go−fishing I−wake−up−late
o
Φ = {Sunday, Holidays}.
Reiter there is only the extension containing Sunday, Holidays, I−wake−up−late (by first applying the second and then the first default)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
60 / 104
Lukaszewicz’s Fishing example Let T = h∆, Φi where ∆= n
Sunday : I−go−fishing∧¬I−wake−up−late Holidays : I−wake−up−late , I−go−fishing I−wake−up−late
o
Φ = {Sunday, Holidays}.
Reiter there is only the extension containing Sunday, Holidays, I−wake−up−late (by first applying the second and then the first default)
Lukaszewicz we also(!) have the extension that is the result of first applying the first default
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
60 / 104
Lukaszewicz’s Fishing example Let T = h∆, Φi where ∆= n
Sunday : I−go−fishing∧¬I−wake−up−late Holidays : I−wake−up−late , I−go−fishing I−wake−up−late
o
Φ = {Sunday, Holidays}.
Reiter there is only the extension containing Sunday, Holidays, I−wake−up−late (by first applying the second and then the first default)
Lukaszewicz we also(!) have the extension that is the result of first applying the first default
Question What do you make of it? Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
60 / 104
Poole’s Lottery Paradox Let T = h∆, Φi where bird(x) : flies(x) ∧ ¬penguin(x) ∆= , flies(x) ∧ ¬penguin(x) bird(x) : treenest(x) ∧ ¬sandpiper(x) ,... treenest(x) ∧ ¬sandpiper(x) Φ = {bird(Tweety)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
61 / 104
Poole’s Lottery Paradox Let T = h∆, Φi where bird(x) : flies(x) ∧ ¬penguin(x) ∆= , flies(x) ∧ ¬penguin(x) bird(x) : treenest(x) ∧ ¬sandpiper(x) ,... treenest(x) ∧ ¬sandpiper(x) Φ = {bird(Tweety)}
Problem However, then we conclude ¬penguin(x) ∧ ¬sandpiper(x) ∧ ¬ . . . for all bird-species. But then Tweety does not belong to any species of birds. Typical birds (in an ideal sense) do not exist.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
61 / 104
Reflections about Extensions should there always be extensions?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
62 / 104
Reflections about Extensions should there always be extensions? what do extensions represent?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
62 / 104
Reflections about Extensions should there always be extensions? what do extensions represent? equilibrium states of a rational reasoner?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
62 / 104
Reflections about Extensions should there always be extensions? what do extensions represent? equilibrium states of a rational reasoner? “different, possibly conflicting conclusion sets as rational outcomes based on initial information” (Horty, 2005)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
62 / 104
Reflections about Extensions should there always be extensions? what do extensions represent? equilibrium states of a rational reasoner? “different, possibly conflicting conclusion sets as rational outcomes based on initial information” (Horty, 2005) good reasons approach: if A is in an extension then there are good reasons to suppose A (see Nixon)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
62 / 104
Reflections about Extensions should there always be extensions? what do extensions represent? equilibrium states of a rational reasoner? “different, possibly conflicting conclusion sets as rational outcomes based on initial information” (Horty, 2005) good reasons approach: if A is in an extension then there are good reasons to suppose A (see Nixon)
are some extensions preferable to others?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
62 / 104
Reflections about Extensions should there always be extensions? what do extensions represent? equilibrium states of a rational reasoner? “different, possibly conflicting conclusion sets as rational outcomes based on initial information” (Horty, 2005) good reasons approach: if A is in an extension then there are good reasons to suppose A (see Nixon)
are some extensions preferable to others? should some extensions be filtered out?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
62 / 104
Reflections about Extensions should there always be extensions? what do extensions represent? equilibrium states of a rational reasoner? “different, possibly conflicting conclusion sets as rational outcomes based on initial information” (Horty, 2005) good reasons approach: if A is in an extension then there are good reasons to suppose A (see Nixon)
are some extensions preferable to others? should some extensions be filtered out? how to build extensions (naturally) in order to explicate actual default reasoning?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
62 / 104
Reflections about Extensions should there always be extensions? what do extensions represent? equilibrium states of a rational reasoner? “different, possibly conflicting conclusion sets as rational outcomes based on initial information” (Horty, 2005) good reasons approach: if A is in an extension then there are good reasons to suppose A (see Nixon)
are some extensions preferable to others? should some extensions be filtered out? how to build extensions (naturally) in order to explicate actual default reasoning? should floating conclusions be accepted?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
62 / 104
Problems with Disjunctions
Let T = h∆, Φi where Quaker(x) : dove(x) republican(x) : hawk(x) , , ∆= dove(x) hawk(x)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
63 / 104
Problems with Disjunctions
Let T = h∆, Φi where Quaker(x) : dove(x) republican(x) : hawk(x) , , ∆= dove(x) hawk(x) Φ = {Quaker(Peter) ∨ republican(Peter), Quaker(Anne) ∨ Quaker(George)}.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
63 / 104
Problems with Disjunctions
Let T = h∆, Φi where Quaker(x) : dove(x) republican(x) : hawk(x) , , ∆= dove(x) hawk(x) Φ = {Quaker(Peter) ∨ republican(Peter), Quaker(Anne) ∨ Quaker(George)}.
Problem we don’t get hawk(Peter) ∨ dove(Peter),
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
63 / 104
Problems with Disjunctions
Let T = h∆, Φi where Quaker(x) : dove(x) republican(x) : hawk(x) , , ∆= dove(x) hawk(x) Φ = {Quaker(Peter) ∨ republican(Peter), Quaker(Anne) ∨ Quaker(George)}.
Problem we don’t get hawk(Peter) ∨ dove(Peter), dove(Anne) ∨ dove(George).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
63 / 104
"Naming defaults"
Let T = h∆, Φi where n o : birdsfly(x) ∆ = > birdsfly(x)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
64 / 104
"Naming defaults"
Let T = h∆, Φi where n o : birdsfly(x) ∆ = > birdsfly(x) Φ consists of
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
64 / 104
"Naming defaults"
Let T = h∆, Φi where n o : birdsfly(x) ∆ = > birdsfly(x) Φ consists of ∀x(birdsfly(x) ∧ bird(x) → flies(x))
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
64 / 104
"Naming defaults"
Let T = h∆, Φi where n o : birdsfly(x) ∆ = > birdsfly(x) Φ consists of ∀x(birdsfly(x) ∧ bird(x) → flies(x)) ∀x(bird(x) ∧ baby(x) → ¬birdsfly(x))
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
64 / 104
"Naming defaults"
Let T = h∆, Φi where n o : birdsfly(x) ∆ = > birdsfly(x) Φ consists of ∀x(birdsfly(x) ∧ bird(x) → flies(x)) ∀x(bird(x) ∧ baby(x) → ¬birdsfly(x)) bird(Tweety), bird(Polly)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
64 / 104
"Naming defaults"
Let T = h∆, Φi where n o : birdsfly(x) ∆ = > birdsfly(x) Φ consists of ∀x(birdsfly(x) ∧ bird(x) → flies(x)) ∀x(bird(x) ∧ baby(x) → ¬birdsfly(x)) bird(Tweety), bird(Polly) bird(Anne) ∨ bird(George)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
64 / 104
"Naming defaults"
Let T = h∆, Φi where n o : birdsfly(x) ∆ = > birdsfly(x) Φ consists of ∀x(birdsfly(x) ∧ bird(x) → flies(x)) ∀x(bird(x) ∧ baby(x) → ¬birdsfly(x)) bird(Tweety), bird(Polly) bird(Anne) ∨ bird(George) baby(Polly), baby(Keith)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
64 / 104
"Naming defaults"
Let T = h∆, Φi where n o : birdsfly(x) ∆ = > birdsfly(x) Φ consists of ∀x(birdsfly(x) ∧ bird(x) → flies(x)) ∀x(bird(x) ∧ baby(x) → ¬birdsfly(x)) bird(Tweety), bird(Polly) bird(Anne) ∨ bird(George) baby(Polly), baby(Keith) ¬flies(Fred)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
64 / 104
"Naming defaults" The good: flies(Anne) ∨ flies(George) Let T = h∆, Φi where n o : birdsfly(x) ∆ = > birdsfly(x)
flies(Tweety)
Φ consists of ∀x(birdsfly(x) ∧ bird(x) → flies(x)) ∀x(bird(x) ∧ baby(x) → ¬birdsfly(x)) bird(Tweety), bird(Polly) bird(Anne) ∨ bird(George) baby(Polly), baby(Keith) ¬flies(Fred)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
64 / 104
"Naming defaults" The good: flies(Anne) ∨ flies(George) Let T = h∆, Φi where n o : birdsfly(x) ∆ = > birdsfly(x) Φ consists of
flies(Tweety)
The bad: But, in some respect this
∀x(birdsfly(x) ∧ bird(x) → flies(x)) ∀x(bird(x) ∧ baby(x) → ¬birdsfly(x)) proposal is too radical: bird(Tweety), bird(Polly) bird(Anne) ∨ bird(George) baby(Polly), baby(Keith) ¬flies(Fred)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
64 / 104
"Naming defaults" The good: flies(Anne) ∨ flies(George) Let T = h∆, Φi where n o : birdsfly(x) ∆ = > birdsfly(x) Φ consists of
flies(Tweety)
The bad: But, in some respect this
∀x(birdsfly(x) ∧ bird(x) → flies(x)) ∀x(bird(x) ∧ baby(x) → ¬birdsfly(x)) proposal is too radical: bird(Tweety), bird(Polly) ¬bird(Keith) bird(Anne) ∨ bird(George) ¬bird(Fred) baby(Polly), baby(Keith) for any ground term ¬flies(Fred)}
t 6= Polly:
birdsfly(t) bird(t) → (flies(t) ∧ ¬baby(t)) Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
64 / 104
A different modelling of the same example Let T = h∆, Φi where
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
65 / 104
A different modelling of the same example Let T = h∆, Φi where o n : birdsfly(x) ∆ = bird(x)birdsfly(x) Φ consists of ∀x(birdsfly(x) → flies(x)) ∀x(baby(x) → ¬birdsfly(x)) bird(Tweety), bird(Polly) bird(Anne) ∨ bird(George) baby(Polly), baby(Keith) ¬flies(Fred)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
65 / 104
A different modelling of the same example Let T = h∆, Φi where o n : birdsfly(x) ∆ = bird(x)birdsfly(x) Φ consists of ∀x(birdsfly(x) → flies(x)) ∀x(baby(x) → ¬birdsfly(x)) bird(Tweety), bird(Polly) bird(Anne) ∨ bird(George) baby(Polly), baby(Keith) ¬flies(Fred)}
Task Try to see why some of the negative —too strong— consequences of the previous slide are avoided in this proposal. Try to see why flies(Anne) ∨ flies(George) is not anymore in any extension. Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
65 / 104
Semi-normal defaults Let T = h∆, Φi where n o flies(x)∧¬baby(x) ∆ = bird(x) : flies(x)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
66 / 104
Semi-normal defaults Let T = h∆, Φi where n o flies(x)∧¬baby(x) ∆ = bird(x) : flies(x) Φ consists of
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
66 / 104
Semi-normal defaults Let T = h∆, Φi where n o flies(x)∧¬baby(x) ∆ = bird(x) : flies(x) Φ consists of bird(Tweety), bird(Pete), bird(Mary)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
66 / 104
Semi-normal defaults Let T = h∆, Φi where n o flies(x)∧¬baby(x) ∆ = bird(x) : flies(x) Φ consists of bird(Tweety), bird(Pete), bird(Mary) baby(Polly)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
66 / 104
Semi-normal defaults Let T = h∆, Φi where n o flies(x)∧¬baby(x) ∆ = bird(x) : flies(x) Φ consists of bird(Tweety), bird(Pete), bird(Mary) baby(Polly) ¬flies(Fred)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
66 / 104
Semi-normal defaults Let T = h∆, Φi where n o flies(x)∧¬baby(x) ∆ = bird(x) : flies(x) Φ consists of bird(Tweety), bird(Pete), bird(Mary) baby(Polly) ¬flies(Fred) bird(Oscar) ∨ bird(Sylvester)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
66 / 104
Semi-normal defaults Let T = h∆, Φi where n o flies(x)∧¬baby(x) ∆ = bird(x) : flies(x) Φ consists of bird(Tweety), bird(Pete), bird(Mary) baby(Polly) ¬flies(Fred) bird(Oscar) ∨ bird(Sylvester) baby(Pete) ∨ baby(Mary)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
66 / 104
Semi-normal defaults Let T = h∆, Φi where n o flies(x)∧¬baby(x) ∆ = bird(x) : flies(x) Φ consists of bird(Tweety), bird(Pete), bird(Mary) baby(Polly) ¬flies(Fred) bird(Oscar) ∨ bird(Sylvester) baby(Pete) ∨ baby(Mary)}
This prevents the conclusion ¬baby(Tweety).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
66 / 104
Semi-normal defaults Let T = h∆, Φi where n o flies(x)∧¬baby(x) ∆ = bird(x) : flies(x) Φ consists of bird(Tweety), bird(Pete), bird(Mary) baby(Polly) ¬flies(Fred) bird(Oscar) ∨ bird(Sylvester) baby(Pete) ∨ baby(Mary)}
This prevents the conclusion ¬baby(Tweety).
Task Try to see what problems we have with disjunctions for this example.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
66 / 104
The expressive power of semi-normal defaults Lukasziewicz writes: Assume, for instance, that on Sundays I usually go fishing, and suppose that you should remain agnostic about my fishing in rainy Sundays. It seems that the only appropriate representation of this situation is to use the following non-normal default: Sunday : I−go−fishing ∧ ¬rain I−go−fishing
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
67 / 104
The expressive power of semi-normal defaults Lukasziewicz writes: Assume, for instance, that on Sundays I usually go fishing, and suppose that you should remain agnostic about my fishing in rainy Sundays. It seems that the only appropriate representation of this situation is to use the following non-normal default: Sunday : I−go−fishing ∧ ¬rain I−go−fishing
Critically evaluated this claim. 1 2
Why is a normal representation of this default suboptimal? Do you agree with L.’s assessment that the proposed non-normal representation is adequate? Suppose your only knowledge is that it is Sunday.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
67 / 104
Semi-normal defaults and the problem of inconsistent assumptions
Let T = h∆, Φi where bird(x) : flies(x) ∧ ¬dead(x) ∆= flies(x) of−ancient−species(x) : fossilised(x) ∧ dead(x) fossilised(x)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
68 / 104
Semi-normal defaults and the problem of inconsistent assumptions
Let T = h∆, Φi where bird(x) : flies(x) ∧ ¬dead(x) ∆= flies(x) of−ancient−species(x) : fossilised(x) ∧ dead(x) fossilised(x) Φ = {bird(Tweety), of−ancient−species(Tweety)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
68 / 104
Semi-normal defaults and the problem of inconsistent assumptions
Let T = h∆, Φi where bird(x) : flies(x) ∧ ¬dead(x) ∆= flies(x) of−ancient−species(x) : fossilised(x) ∧ dead(x) fossilised(x) Φ = {bird(Tweety), of−ancient−species(Tweety)}
Task Try to see what’s the problem here.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
68 / 104
Semi-normal vs. normal defaults
Compare has−motive(x) : guilty(x) ∧ suspect(x) suspect(x) with has−motive(x) : guilty(x) ∧ suspect(x) guilty(x) ∧ suspect(x)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
69 / 104
Semi-Normal defaults ala Brewka/Levesque Let T = h∆, Φi where n o > : flies(x) ∆ = bird(x)→flies(x)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
70 / 104
Semi-Normal defaults ala Brewka/Levesque Let T = h∆, Φi where n o > : flies(x) ∆ = bird(x)→flies(x) Φ consists of
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
70 / 104
Semi-Normal defaults ala Brewka/Levesque Let T = h∆, Φi where n o > : flies(x) ∆ = bird(x)→flies(x) Φ consists of bird(Tweety), bird(Polly)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
70 / 104
Semi-Normal defaults ala Brewka/Levesque Let T = h∆, Φi where n o > : flies(x) ∆ = bird(x)→flies(x) Φ consists of bird(Tweety), bird(Polly) baby(Polly), ¬flies(Fred)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
70 / 104
Semi-Normal defaults ala Brewka/Levesque Let T = h∆, Φi where n o > : flies(x) ∆ = bird(x)→flies(x) Φ consists of bird(Tweety), bird(Polly) baby(Polly), ¬flies(Fred) bird(Oscar) ∨ bird(Sylvester)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
70 / 104
Semi-Normal defaults ala Brewka/Levesque Let T = h∆, Φi where n o > : flies(x) ∆ = bird(x)→flies(x) Φ consists of bird(Tweety), bird(Polly) baby(Polly), ¬flies(Fred) bird(Oscar) ∨ bird(Sylvester) ¬flies(Pete) ∨ ¬flies(Mary)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
70 / 104
Semi-Normal defaults ala Brewka/Levesque Let T = h∆, Φi where n o > : flies(x) ∆ = bird(x)→flies(x) Φ consists of bird(Tweety), bird(Polly) baby(Polly), ¬flies(Fred) bird(Oscar) ∨ bird(Sylvester) ¬flies(Pete) ∨ ¬flies(Mary)}
Task We cannot conclude ¬bird(Fred). See why.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
70 / 104
Semi-Normal defaults ala Brewka/Levesque Let T = h∆, Φi where n o > : flies(x) ∆ = bird(x)→flies(x) Φ consists of bird(Tweety), bird(Polly) baby(Polly), ¬flies(Fred) bird(Oscar) ∨ bird(Sylvester) ¬flies(Pete) ∨ ¬flies(Mary)}
Task We cannot conclude ¬bird(Fred). See why. Observe what happens to Pete and Mary.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
70 / 104
The Finch-Example
Let T = h∆, n Φi where : green−island(x) ∆ = ruffed−finch(x) , green−island(x) least−ruffed−finch(x) : green−island(x)∨sand−island(x) green−island(x)∨sand−island(x)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
o
September 2, 2015
71 / 104
The Finch-Example
Let T = h∆, n Φi where : green−island(x) ∆ = ruffed−finch(x) , green−island(x) least−ruffed−finch(x) : green−island(x)∨sand−island(x) green−island(x)∨sand−island(x)
o
Φ consists of
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
71 / 104
The Finch-Example
Let T = h∆, n Φi where : green−island(x) ∆ = ruffed−finch(x) , green−island(x) least−ruffed−finch(x) : green−island(x)∨sand−island(x) green−island(x)∨sand−island(x)
o
Φ consists of least−ruffed−finch(Frank)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
71 / 104
The Finch-Example
Let T = h∆, n Φi where : green−island(x) ∆ = ruffed−finch(x) , green−island(x) least−ruffed−finch(x) : green−island(x)∨sand−island(x) green−island(x)∨sand−island(x)
o
Φ consists of least−ruffed−finch(Frank) ∀x(least−ruffed−finch(x) → ruffed−finch(x))}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
71 / 104
The Finch-Example
Let T = h∆, n Φi where : green−island(x) ∆ = ruffed−finch(x) , green−island(x) least−ruffed−finch(x) : green−island(x)∨sand−island(x) green−island(x)∨sand−island(x)
o
Φ consists of least−ruffed−finch(Frank) ∀x(least−ruffed−finch(x) → ruffed−finch(x))}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
71 / 104
The Finch-Example
Let T = h∆, n Φi where : green−island(x) ∆ = ruffed−finch(x) , green−island(x) least−ruffed−finch(x) : green−island(x)∨sand−island(x) green−island(x)∨sand−island(x)
o
Φ consists of least−ruffed−finch(Frank) ∀x(least−ruffed−finch(x) → ruffed−finch(x))}
Problem
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
71 / 104
The Finch-Example
Let T = h∆, n Φi where : green−island(x) ∆ = ruffed−finch(x) , green−island(x) least−ruffed−finch(x) : green−island(x)∨sand−island(x) green−island(x)∨sand−island(x)
o
Φ consists of least−ruffed−finch(Frank) ∀x(least−ruffed−finch(x) → ruffed−finch(x))}
Problem the unique extension includes both green−island(Frank) and green−island(Frank) ∨ sand−island(Frank) (since both defaults are triggered) Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
71 / 104
Default logic and monotonicity Nonmonotonicity, both in the set of defaults ∆ in the set of facts Φ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
72 / 104
Default logic and monotonicity Nonmonotonicity, both in the set of defaults ∆ in the set of facts Φ
Not even cautious monotonic Here’s an example that goes back to Makinson:
∆=
n
> : p p∨q : ¬p , p ¬p
o
Φ1 = ∅ Φ2 = {p ∨ q} Check, what happens! Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
72 / 104
The problem with ’negative cycles’
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
73 / 104
The problem with ’negative cycles’
for a way to deal with this problem see Antonelli (1999)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
73 / 104
Prioritized default theories (Horty (2007, 2012))
hΦ, ∆, ≺i where we have a strict partial order on the defaults: ≺. This represents a priority/preference relation. Depending on the application this may indicate the rank of an authority from which the information stems, the reliability of the source, specificity relations, etc.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
74 / 104
Prioritized default theories (Horty (2007, 2012))
hΦ, ∆, ≺i where we have a strict partial order on the defaults: ≺. This represents a priority/preference relation. Depending on the application this may indicate the rank of an authority from which the information stems, the reliability of the source, specificity relations, etc. A strict partial order is (i) irreflexive (A 6< A), (ii) asymmetric (If A < B then B 6< A), and (iii) transitive (A < B and B < C implies A < C ). Graphically they are represented by directed acyclic graphs (the transitive closure is usually not represented).
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
74 / 104
Scenarios vs. extensions
Given an ordered default theory hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
75 / 104
Scenarios vs. extensions
Given an ordered default theory hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
75 / 104
Proper Scenarios
like in the non-prioritised case, not just any superset of Φ constitutes an extension, we are also now interested in scenarios that in some sense represent the set of defaults a rational agent would select/use given Φ and <.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
76 / 104
Proper Scenarios
like in the non-prioritised case, not just any superset of Φ constitutes an extension, we are also now interested in scenarios that in some sense represent the set of defaults a rational agent would select/use given Φ and <. E.g., a scenario should not generate conflicting beliefs.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
76 / 104
Proper Scenarios
like in the non-prioritised case, not just any superset of Φ constitutes an extension, we are also now interested in scenarios that in some sense represent the set of defaults a rational agent would select/use given Φ and <. E.g., a scenario should not generate conflicting beliefs. Moreover, priorities should be taken into account.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
76 / 104
Proper Scenarios
like in the non-prioritised case, not just any superset of Φ constitutes an extension, we are also now interested in scenarios that in some sense represent the set of defaults a rational agent would select/use given Φ and <. E.g., a scenario should not generate conflicting beliefs. Moreover, priorities should be taken into account. Hence, we are interested in what Horty calls proper scenarios and the belief sets generated by them.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
76 / 104
The idea is again to build up scenarios stepwise similar as in the procedural approaches to build extensions.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
77 / 104
The idea is again to build up scenarios stepwise similar as in the procedural approaches to build extensions. we start with our facts Φ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
77 / 104
The idea is again to build up scenarios stepwise similar as in the procedural approaches to build extensions. we start with our facts Φ What are interesting defaults to take into account?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
77 / 104
The idea is again to build up scenarios stepwise similar as in the procedural approaches to build extensions. we start with our facts Φ What are interesting defaults to take into account? we have the choice between triggered defaults: Triggered(S) = {δ ∈ ∆ | Φ ∪ Conclusion(S) ` Premise(δ)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
77 / 104
The idea is again to build up scenarios stepwise similar as in the procedural approaches to build extensions. we start with our facts Φ What are interesting defaults to take into account? we have the choice between triggered defaults: Triggered(S) = {δ ∈ ∆ | Φ ∪ Conclusion(S) ` Premise(δ)} However, some triggered defaults may bad candidates since they (i.e., their conclusions) conflict with our belief set. Hence, we want to neglect: Conflicted(S) = {δ ∈ ∆ | Φ ∪ Conclusion(S) ` ¬Conclusion(δ)}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
77 / 104
What about the priorities? Also, we have to take into account our priorities.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
78 / 104
What about the priorities? Also, we have to take into account our priorities. A first idea would be to pick one of the highest ranked triggered defaults.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
78 / 104
What about the priorities? Also, we have to take into account our priorities. A first idea would be to pick one of the highest ranked triggered defaults. this is not Horty’s approach (see later: the order puzzle)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
78 / 104
What about the priorities? Also, we have to take into account our priorities. A first idea would be to pick one of the highest ranked triggered defaults. this is not Horty’s approach (see later: the order puzzle)
We also have to take into account that sometimes sets of defaults “defeat” other defaults
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
78 / 104
What about the priorities? Also, we have to take into account our priorities. A first idea would be to pick one of the highest ranked triggered defaults. this is not Horty’s approach (see later: the order puzzle)
We also have to take into account that sometimes sets of defaults “defeat” other defaults We write S < S 0 where S, S 0 ⊆ ∆ and δ < δ 0 for all δ ∈ S and all δ0 ∈ S 0
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
78 / 104
What about the priorities? Also, we have to take into account our priorities. A first idea would be to pick one of the highest ranked triggered defaults. this is not Horty’s approach (see later: the order puzzle)
We also have to take into account that sometimes sets of defaults “defeat” other defaults We write S < S 0 where S, S 0 ⊆ ∆ and δ < δ 0 for all δ ∈ S and all δ0 ∈ S 0
Example E.g., where δ1 = a → b, δ2 = b → c and δ3 = a → ¬c and δ3 < δ1 , δ2 , we have {δ3 } < {δ1 , δ2 } Note that {a} ∪ Conclusion({δ1 , δ2 }) ` ¬Conclusion(δ3 ) in this sense {δ1 , δ2 } defeats δ3 Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
78 / 104
Horty makes the idea above precise relative to a given scenario S.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
79 / 104
Horty makes the idea above precise relative to a given scenario S. The set Defeated(S) is the set of all δ ∈ ∆ such that there is a ∆d ⊆ Triggered(S) (a defeating set) for which
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
79 / 104
Horty makes the idea above precise relative to a given scenario S. The set Defeated(S) is the set of all δ ∈ ∆ such that there is a ∆d ⊆ Triggered(S) (a defeating set) for which {δ} < ∆d and
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
79 / 104
Horty makes the idea above precise relative to a given scenario S. The set Defeated(S) is the set of all δ ∈ ∆ such that there is a ∆d ⊆ Triggered(S) (a defeating set) for which {δ} < ∆d and there is a Sa ⊆ S (an accommodation set) for which
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
79 / 104
Horty makes the idea above precise relative to a given scenario S. The set Defeated(S) is the set of all δ ∈ ∆ such that there is a ∆d ⊆ Triggered(S) (a defeating set) for which {δ} < ∆d and there is a Sa ⊆ S (an accommodation set) for which 1
Sa < ∆d
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
79 / 104
Horty makes the idea above precise relative to a given scenario S. The set Defeated(S) is the set of all δ ∈ ∆ such that there is a ∆d ⊆ Triggered(S) (a defeating set) for which {δ} < ∆d and there is a Sa ⊆ S (an accommodation set) for which 1 2
Sa < ∆d Φ ∪ Conclusion((S \ Sa ) ∪ ∆d ) is consistent
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
79 / 104
Horty makes the idea above precise relative to a given scenario S. The set Defeated(S) is the set of all δ ∈ ∆ such that there is a ∆d ⊆ Triggered(S) (a defeating set) for which {δ} < ∆d and there is a Sa ⊆ S (an accommodation set) for which 1 2 3
Sa < ∆d Φ ∪ Conclusion((S \ Sa ) ∪ ∆d ) is consistent Φ ∪ Conclusion((S \ Sa ) ∪ ∆d ) ` ¬Conclusion(δ)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
79 / 104
Some examples
Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Suppose S = {B → F }.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
80 / 104
Some examples
Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Suppose S = {B → F }. P → ¬F ∈ Triggered(S)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
80 / 104
Some examples
Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Suppose S = {B → F }. P → ¬F ∈ Triggered(S) {B → F } is an accommodation set for the defeating set {P → ¬F } w.r.t. B → F
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
80 / 104
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Φ = {a}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Let S1 = {a → ¬c}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Let S1 = {a → ¬c} a → b is triggered by S
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Let S1 = {a → ¬c} a → b is triggered by S no defeating takes place
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Let S1 = {a → ¬c} a → b is triggered by S no defeating takes place
Let S2 = {a → ¬c, a → b}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Let S1 = {a → ¬c} a → b is triggered by S no defeating takes place
Let S2 = {a → ¬c, a → b} b → c is triggered
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Let S1 = {a → ¬c} a → b is triggered by S no defeating takes place
Let S2 = {a → ¬c, a → b} b → c is triggered b → c is conflicted by S
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Let S1 = {a → ¬c} a → b is triggered by S no defeating takes place
Let S2 = {a → ¬c, a → b} b → c is triggered b → c is conflicted by S {a → ¬c} is defeated by S
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Let S1 = {a → ¬c} a → b is triggered by S no defeating takes place
Let S2 = {a → ¬c, a → b} b → c is triggered b → c is conflicted by S {a → ¬c} is defeated by S note that {a → ¬c} is an accommodation set for the defeating set {b → c} Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
81 / 104
An intuitive problem?
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
82 / 104
An intuitive problem?
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
82 / 104
An intuitive problem?
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
82 / 104
An intuitive problem?
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
82 / 104
An intuitive problem?
Take T = hΦ, ∆,
Φ = {a}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
82 / 104
An intuitive problem?
Take T = hΦ, ∆,
Φ = {a} δ1 < δ3
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
82 / 104
An intuitive problem?
Take T = hΦ, ∆,
Φ = {a} δ1 < δ3
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
82 / 104
An intuitive problem?
Take T = hΦ, ∆,
Φ = {a} δ1 < δ3
Take S = {a → b, b → c}
Note that a → b ∈ Defeated(S) {a → b} is an accommodation set for the defeating set {c → ¬b} Note that Φ ∪ Conclusion(b → c, c → ¬b) ` ¬b However, neither b → c nor c → ¬b is triggered w.r.t. Φ!
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
82 / 104
Horty’s procedure for proper scenarios and extensions Given an ordered default theory hΦ, ∆,
guess S
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
83 / 104
Horty’s procedure for proper scenarios and extensions Given an ordered default theory hΦ, ∆,
guess S
2
the initial scenario is S0 = ∅
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
83 / 104
Horty’s procedure for proper scenarios and extensions Given an ordered default theory hΦ, ∆,
guess S the initial scenario is S0 = ∅ do the following until a fixed point is reached
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
83 / 104
Horty’s procedure for proper scenarios and extensions Given an ordered default theory hΦ, ∆,
guess S the initial scenario is S0 = ∅ do the following until a fixed point is reached 1
Si+1 : add all δ ∈ ∆ to Si that satisfy the following conditions
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
83 / 104
Horty’s procedure for proper scenarios and extensions Given an ordered default theory hΦ, ∆,
guess S the initial scenario is S0 = ∅ do the following until a fixed point is reached 1
Si+1 : add all δ ∈ ∆ to Si that satisfy the following conditions 1
δ ∈ Triggered(Si )
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
83 / 104
Horty’s procedure for proper scenarios and extensions Given an ordered default theory hΦ, ∆,
guess S the initial scenario is S0 = ∅ do the following until a fixed point is reached 1
Si+1 : add all δ ∈ ∆ to Si that satisfy the following conditions 1 2
δ ∈ Triggered(Si ) δ∈ / Conflicted(S) (here you need to make use of your guess!)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
83 / 104
Horty’s procedure for proper scenarios and extensions Given an ordered default theory hΦ, ∆,
guess S the initial scenario is S0 = ∅ do the following until a fixed point is reached 1
Si+1 : add all δ ∈ ∆ to Si that satisfy the following conditions 1 2 3
δ ∈ Triggered(Si ) δ∈ / Conflicted(S) (here you need to make use of your guess!) δ∈ / Defeated(S) (also here)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
83 / 104
Horty’s procedure for proper scenarios and extensions Given an ordered default theory hΦ, ∆,
guess S the initial scenario is S0 = ∅ do the following until a fixed point is reached 1
Si+1 : add all δ ∈ ∆ to Si that satisfy the following conditions 1 2 3
4
δ ∈ Triggered(Si ) δ∈ / Conflicted(S) (here you need to make use of your guess!) δ∈ / Defeated(S) (also here)
let your fixed point be S 0 .
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
83 / 104
Horty’s procedure for proper scenarios and extensions Given an ordered default theory hΦ, ∆,
guess S the initial scenario is S0 = ∅ do the following until a fixed point is reached 1
Si+1 : add all δ ∈ ∆ to Si that satisfy the following conditions 1 2 3
4
δ ∈ Triggered(Si ) δ∈ / Conflicted(S) (here you need to make use of your guess!) δ∈ / Defeated(S) (also here)
let your fixed point be S 0 . 1
If S = S 0 you’re done. Then S is a proper scenario and Ξ = Cn(Φ ∪ Conclusion(S)) is an extension.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
83 / 104
Horty’s procedure for proper scenarios and extensions Given an ordered default theory hΦ, ∆,
guess S the initial scenario is S0 = ∅ do the following until a fixed point is reached 1
Si+1 : add all δ ∈ ∆ to Si that satisfy the following conditions 1 2 3
4
δ ∈ Triggered(Si ) δ∈ / Conflicted(S) (here you need to make use of your guess!) δ∈ / Defeated(S) (also here)
let your fixed point be S 0 . 1
2
If S = S 0 you’re done. Then S is a proper scenario and Ξ = Cn(Φ ∪ Conclusion(S)) is an extension. Otherwise, start anew with another guess.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
83 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round: S0 = ∅
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 )
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) P → ¬F ∈ Conflicted(S)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) P → ¬F ∈ Conflicted(S) but: B → F ∈ Defeated(S)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) P → ¬F ∈ Conflicted(S) but: B → F ∈ Defeated(S) nothing is added!
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) P → ¬F ∈ Conflicted(S) but: B → F ∈ Defeated(S) nothing is added!
Guess: S = {P → ¬F }
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) P → ¬F ∈ Conflicted(S) but: B → F ∈ Defeated(S) nothing is added!
Guess: S = {P → ¬F } First round:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) P → ¬F ∈ Conflicted(S) but: B → F ∈ Defeated(S) nothing is added!
Guess: S = {P → ¬F } First round: S0 = ∅
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) P → ¬F ∈ Conflicted(S) but: B → F ∈ Defeated(S) nothing is added!
Guess: S = {P → ¬F } First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 )
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) P → ¬F ∈ Conflicted(S) but: B → F ∈ Defeated(S) nothing is added!
Guess: S = {P → ¬F } First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) B → F ∈ Conflicted(S)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) P → ¬F ∈ Conflicted(S) but: B → F ∈ Defeated(S) nothing is added!
Guess: S = {P → ¬F } First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) B → F ∈ Conflicted(S) P → ¬F ∈ / Defeated(S)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) P → ¬F ∈ Conflicted(S) but: B → F ∈ Defeated(S) nothing is added!
Guess: S = {P → ¬F } First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) B → F ∈ Conflicted(S) P → ¬F ∈ / Defeated(S) hence: S1 = {P → ¬F }
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
An example Take T = h{P, P ⊃ B}, {δ1 , δ2 }, {(δ1 , δ2 )}i where δ1 = B → F and δ2 = P → ¬F . Guess S = {B → F }. First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) P → ¬F ∈ Conflicted(S) but: B → F ∈ Defeated(S) nothing is added!
Guess: S = {P → ¬F } First round:
S0 = ∅ B → F , P → ¬F ∈ Triggered(S0 ) B → F ∈ Conflicted(S) P → ¬F ∈ / Defeated(S) hence: S1 = {P → ¬F }
second round: fixed point reached. Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
84 / 104
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Φ = {a}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Guess: S = {a → ¬c}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Guess: S = {a → ¬c} Round 1: S0 = ∅
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Guess: S = {a → ¬c} Round 1: S0 = ∅ Triggered: a → b and a → ¬c
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Guess: S = {a → ¬c} Round 1: S0 = ∅ Triggered: a → b and a → ¬c a→b∈ / Defeated(S)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Guess: S = {a → ¬c} Round 1: S0 = ∅ Triggered: a → b and a → ¬c a→b∈ / Defeated(S) a→b∈ / Conflicted(S)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Guess: S = {a → ¬c} Round 1: S0 = ∅ Triggered: a → b and a → ¬c a→b∈ / Defeated(S) a→b∈ / Conflicted(S) hence, a → b has to be added!
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Guess: S = {a → ¬c} Round 1: S0 = ∅ Triggered: a → b and a → ¬c a→b∈ / Defeated(S) a→b∈ / Conflicted(S) hence, a → b has to be added!
Guess: S = {a → b, b → c}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Take T = hΦ, ∆,
Φ = {a} δ3 < δ1 and δ3 < δ2 Guess: S = {a → ¬c} Round 1: S0 = ∅ Triggered: a → b and a → ¬c a→b∈ / Defeated(S) a→b∈ / Conflicted(S) hence, a → b has to be added!
Guess: S = {a → b, b → c} This works: check out why
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
85 / 104
Theorem Where hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
86 / 104
Theorem Where hΦ, ∆,
Theorem Where T< = hΦ, ∆,
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
86 / 104
Another problematic example: The Order Puzzle Let T = hΦ, ∆, → A δ2 = > → B δ3 = A → ¬B
Φ=∅ δ1 < δ2 < δ3
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
87 / 104
Another problematic example: The Order Puzzle Let T = hΦ, ∆, → A δ2 = > → B δ3 = A → ¬B
Φ=∅ δ1 < δ2 < δ3
We have one proper scenarios:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
87 / 104
Another problematic example: The Order Puzzle Let T = hΦ, ∆, → A δ2 = > → B δ3 = A → ¬B
Φ=∅ δ1 < δ2 < δ3
We have one proper scenarios: S1 = {δ1 , δ2 }? Nope!
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
87 / 104
Another problematic example: The Order Puzzle Let T = hΦ, ∆, → A δ2 = > → B δ3 = A → ¬B
Φ=∅ δ1 < δ2 < δ3
We have one proper scenarios: S1 = {δ1 , δ2 }? Nope! S2 = {δ1 , δ3 }
What do you think? Interpret the premises in terms of conditional commands, and in terms of various pieces of information with different degrees of reliability. Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
87 / 104
The normative reading by Horty (p.392, 2007)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
88 / 104
A similar problem Let RC = “being a resident of Cuba”; RN = “being a resident of North-America; CC = “being a citizen of Cuba”; CU = “being a citizen of the USA”; and VU = “having voting rights in the USA”. Let T = hΦ, ∆,
Φ = {RC , RC ⊃ RN, ¬(CC ∧ CU), ¬(CC ∧ VU)} δ1 < δ2 < δ3
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
89 / 104
A similar problem Let RC = “being a resident of Cuba”; RN = “being a resident of North-America; CC = “being a citizen of Cuba”; CU = “being a citizen of the USA”; and VU = “having voting rights in the USA”. Let T = hΦ, ∆,
Φ = {RC , RC ⊃ RN, ¬(CC ∧ CU), ¬(CC ∧ VU)} δ1 < δ2 < δ3
There are two proper scenarios: S1 = {δ2 } and S2 = {δ1 , δ3 }. (this is problematic) Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
89 / 104
An intuitive problem?
Take T = hΦ, ∆,
Φ = {a} δ1 < δ3
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
90 / 104
An intuitive problem?
Take T = hΦ, ∆,
Φ = {a} δ1 < δ3
No extension/proper scenario!
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
90 / 104
Multiple occurrences of the same defaults Let T = hΦ, ∆,
=>→A = > → ¬A =>→A = > → ¬A
Φ=∅ δ1 < δ2 and δ3 < δ4
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
91 / 104
Multiple occurrences of the same defaults Let T = hΦ, ∆,
=>→A = > → ¬A =>→A = > → ¬A
Φ=∅ δ1 < δ2 and δ3 < δ4
We have two proper scenarios: S1 = {δ2 , δ4 } S2 = {δ1 , δ3 } (problematic!)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
91 / 104
Multiple occurrences of the same defaults Let T = hΦ, ∆,
=>→A = > → ¬A =>→A = > → ¬A
Φ=∅ δ1 < δ2 and δ3 < δ4
We have two proper scenarios: S1 = {δ2 , δ4 } S2 = {δ1 , δ3 } (problematic!)
Solution . . . redefine < for sets of defaults? Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
91 / 104
Some philosophical Questions in the end
What should we derive in default logic? What is our standard of adequacy, what is our ideal rational agent?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
92 / 104
Some philosophical Questions in the end
What should we derive in default logic? What is our standard of adequacy, what is our ideal rational agent? How descriptive of actual reasoning should default logic be, and how normative should it be?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
92 / 104
Some philosophical Questions in the end
What should we derive in default logic? What is our standard of adequacy, what is our ideal rational agent? How descriptive of actual reasoning should default logic be, and how normative should it be? What are the “core properties” of default reasoning?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
92 / 104
Looking back . . . : what did we learn?
What is a default theory?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
93 / 104
Looking back . . . : what did we learn?
What is a default theory? We learned about two different notions for extensions.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
93 / 104
Looking back . . . : what did we learn?
What is a default theory? We learned about two different notions for extensions. There may be multiple extensions, for Reiter’s logic there may be none.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
93 / 104
Looking back . . . : what did we learn?
What is a default theory? We learned about two different notions for extensions. There may be multiple extensions, for Reiter’s logic there may be none. We have highlighted various problems for default logic:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
93 / 104
Looking back . . . : what did we learn?
What is a default theory? We learned about two different notions for extensions. There may be multiple extensions, for Reiter’s logic there may be none. We have highlighted various problems for default logic: specificity
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
93 / 104
Looking back . . . : what did we learn?
What is a default theory? We learned about two different notions for extensions. There may be multiple extensions, for Reiter’s logic there may be none. We have highlighted various problems for default logic: specificity disjunctive facts
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
93 / 104
Looking back . . . : what did we learn?
What is a default theory? We learned about two different notions for extensions. There may be multiple extensions, for Reiter’s logic there may be none. We have highlighted various problems for default logic: specificity disjunctive facts disjunctive exceptions
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
93 / 104
Looking back . . . : what did we learn?
What is a default theory? We learned about two different notions for extensions. There may be multiple extensions, for Reiter’s logic there may be none. We have highlighted various problems for default logic: specificity disjunctive facts disjunctive exceptions consistency of assumptions
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
93 / 104
Looking back . . . : what did we learn?
What is a default theory? We learned about two different notions for extensions. There may be multiple extensions, for Reiter’s logic there may be none. We have highlighted various problems for default logic: specificity disjunctive facts disjunctive exceptions consistency of assumptions etc.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
93 / 104
Looking back . . . : what did we learn?
What is a default theory? We learned about two different notions for extensions. There may be multiple extensions, for Reiter’s logic there may be none. We have highlighted various problems for default logic: specificity disjunctive facts disjunctive exceptions consistency of assumptions etc.
There are various ways of phrasing defaults (normal vs. semi-normal vs. naming defaults, etc.)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
93 / 104
Topic 1
2 3
4 5
Defeasible Reasoning – Some Basic Concepts Some examples to warm up Some conceptual distinctions Nonmonotonic Logic in Context The Dynamics of Defeasible Reasoning Default Logic (in the tradition of Reiter) Warming up Defaults and Default Theories Inferring Alternatives and more examples Meta-Properties Introducing Priorities Summing up Autoepistemic Logic Bibliography Bibliography
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
94 / 104
Idea (Moore (1984, 1985))
Logic to model the reasoning of an (ideal) agent reflecting about her beliefs
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
95 / 104
Idea (Moore (1984, 1985))
Logic to model the reasoning of an (ideal) agent reflecting about her beliefs
Autoepistemic Theory features the beliefs A of an epistemic agent
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
95 / 104
Idea (Moore (1984, 1985))
Logic to model the reasoning of an (ideal) agent reflecting about her beliefs
Autoepistemic Theory features the beliefs A of an epistemic agent including her introspective beliefs about her beliefs BA (and lack thereof)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
95 / 104
Idea (Moore (1984, 1985))
Logic to model the reasoning of an (ideal) agent reflecting about her beliefs
Autoepistemic Theory features the beliefs A of an epistemic agent including her introspective beliefs about her beliefs BA (and lack thereof) closed under rationality constraints (next slide!)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
95 / 104
Stable Autoepistemic Theories
Stalnaker (1993), stability criteria on a autoepistemic theory Γ 1
A ∈ Γ implies BA ∈ Γ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
96 / 104
Stable Autoepistemic Theories
Stalnaker (1993), stability criteria on a autoepistemic theory Γ 1
A ∈ Γ implies BA ∈ Γ
2
A∈ / Γ implies ¬BA ∈ Γ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
96 / 104
Stable Autoepistemic Theories
Stalnaker (1993), stability criteria on a autoepistemic theory Γ 1
A ∈ Γ implies BA ∈ Γ
2
A∈ / Γ implies ¬BA ∈ Γ.
3
classical closure: Cn(Γ) = Γ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
96 / 104
Stable Autoepistemic Theories
Stalnaker (1993), stability criteria on a autoepistemic theory Γ 1
A ∈ Γ implies BA ∈ Γ
2
A∈ / Γ implies ¬BA ∈ Γ.
3
classical closure: Cn(Γ) = Γ.
If Γ is consistent, this implies . . .
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
96 / 104
Stable Autoepistemic Theories
Stalnaker (1993), stability criteria on a autoepistemic theory Γ 1
A ∈ Γ implies BA ∈ Γ
2
A∈ / Γ implies ¬BA ∈ Γ.
3
classical closure: Cn(Γ) = Γ.
If Γ is consistent, this implies . . . 1
A ∈ Γ iff BA ∈ Γ
2
A∈ / Γ iff ¬BA.
You see why?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
96 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which? Ξ1 containing ¬brother Can you see what else is contained?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which? Ξ1 containing ¬brother Can you see what else is contained? by (1), Bi ¬brother ∈ Ξ1
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which? Ξ1 containing ¬brother Can you see what else is contained? by (1), Bi ¬brother ∈ Ξ1 by cons., brother ∈ / Ξ1
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which? Ξ1 containing ¬brother Can you see what else is contained? by (1), Bi ¬brother ∈ Ξ1 by cons., brother ∈ / Ξ1 by (2), ¬Bbrother ∈ Ξ1
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which? Ξ1 containing ¬brother Can you see what else is contained? by (1), Bi ¬brother ∈ Ξ1 by cons., brother ∈ / Ξ1 by (2), ¬Bbrother ∈ Ξ1 by (1), Bi ¬Bbrother ∈ Ξ1
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which? Ξ1 containing ¬brother Can you see what else is contained? by (1), Bi ¬brother ∈ Ξ1
Ξ2 containing brother Can you see what else is contained?
by cons., brother ∈ / Ξ1 by (2), ¬Bbrother ∈ Ξ1 by (1), Bi ¬Bbrother ∈ Ξ1
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which? Ξ1 containing ¬brother Can you see what else is contained? by (1), Bi ¬brother ∈ Ξ1 by cons., brother ∈ / Ξ1
Ξ2 containing brother Can you see what else is contained? by (1), Bi brother ∈ Ξ2
by (2), ¬Bbrother ∈ Ξ1 by (1), Bi ¬Bbrother ∈ Ξ1
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which? Ξ1 containing ¬brother Can you see what else is contained? by (1), Bi ¬brother ∈ Ξ1 by cons., brother ∈ / Ξ1 by (2), ¬Bbrother ∈ Ξ1
Ξ2 containing brother Can you see what else is contained? by (1), Bi brother ∈ Ξ2 by cons., ¬brother ∈ / Ξ2
by (1), Bi ¬Bbrother ∈ Ξ1
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which? Ξ1 containing ¬brother Can you see what else is contained? by (1), Bi ¬brother ∈ Ξ1 by cons., brother ∈ / Ξ1 by (2), ¬Bbrother ∈ Ξ1 by (1), Bi ¬Bbrother ∈ Ξ1
Christian Straßer (RUB, UGENT)
Ξ2 containing brother Can you see what else is contained? by (1), Bi brother ∈ Ξ2 by cons., ¬brother ∈ / Ξ2 by (2), ¬B¬brother ∈ Ξ2
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which? Ξ1 containing ¬brother Can you see what else is contained? by (1), Bi ¬brother ∈ Ξ1 by cons., brother ∈ / Ξ1 by (2), ¬Bbrother ∈ Ξ1 by (1), Bi ¬Bbrother ∈ Ξ1
Ξ2 containing brother Can you see what else is contained? by (1), Bi brother ∈ Ξ2 by cons., ¬brother ∈ / Ξ2 by (2), ¬B¬brother ∈ Ξ2
Do you see a problem?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Stability is not enough . . . Groundedness Take Ξ = {¬Bbrother ⊃ ¬brother}.
Two consistent stable autoepistemic supersets of Ξ: do you see which? Ξ1 containing ¬brother Can you see what else is contained? by (1), Bi ¬brother ∈ Ξ1 by cons., brother ∈ / Ξ1 by (2), ¬Bbrother ∈ Ξ1 by (1), Bi ¬Bbrother ∈ Ξ1
Ξ2 containing brother Can you see what else is contained? by (1), Bi brother ∈ Ξ2 by cons., ¬brother ∈ / Ξ2 by (2), ¬B¬brother ∈ Ξ2
Do you see a problem? The belief in brother in Ξ2 is not grounded given Ξ! Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
97 / 104
Grounding Autoepistemic Theories Here’s how it’s done:
Autoepistemic Extension of Γ Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ}) where Ξ = {A | A ∈ / Γ} BΞ = {BA | A ∈ Ξ}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
98 / 104
Grounding Autoepistemic Theories Here’s how it’s done:
Autoepistemic Extension of Γ Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ}) where Ξ = {A | A ∈ / Γ} BΞ = {BA | A ∈ Ξ} Do you see where there is no autoepistemic extension of Γ = {¬Bbrother ⊃ ¬brother} that contains brother?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
98 / 104
Grounding Autoepistemic Theories Here’s how it’s done:
Autoepistemic Extension of Γ Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ}) where Ξ = {A | A ∈ / Γ} BΞ = {BA | A ∈ Ξ} Do you see where there is no autoepistemic extension of Γ = {¬Bbrother ⊃ ¬brother} that contains brother? Suppose there were one Ξ! But then Γ ∪ BΞ ∪ ¬BΞ ` brother — but this is impossible!
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
98 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Take Γ = {¬BA ⊃ A}.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Take Γ = {¬BA ⊃ A}. Let Ξ be any classically closed superset of Γ.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Take Γ = {¬BA ⊃ A}. Let Ξ be any classically closed superset of Γ. suppose A ∈ Ξ:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Take Γ = {¬BA ⊃ A}. Let Ξ be any classically closed superset of Γ. suppose A ∈ Ξ: but how to ground A?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Take Γ = {¬BA ⊃ A}. Let Ξ be any classically closed superset of Γ. suppose A ∈ Ξ: but how to ground A? in view of Γ the only way is to have also ¬BA
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Take Γ = {¬BA ⊃ A}. Let Ξ be any classically closed superset of Γ. suppose A ∈ Ξ: but how to ground A? in view of Γ the only way is to have also ¬BA by we have BA ∈ Ξ
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Take Γ = {¬BA ⊃ A}. Let Ξ be any classically closed superset of Γ. suppose A ∈ Ξ: but how to ground A? in view of Γ the only way is to have also ¬BA by we have BA ∈ Ξ hence, Ξ is trivial (by classical closure)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Take Γ = {¬BA ⊃ A}. Let Ξ be any classically closed superset of Γ. suppose A ∈ Ξ: but how to ground A? in view of Γ the only way is to have also ¬BA by we have BA ∈ Ξ hence, Ξ is trivial (by classical closure) but then BΞ = ∅, a contradiction, since now there is no way to derive ¬BA
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Take Γ = {¬BA ⊃ A}. Let Ξ be any classically closed superset of Γ. suppose A ∈ Ξ: but how to ground A? in view of Γ the only way is to have also ¬BA by we have BA ∈ Ξ hence, Ξ is trivial (by classical closure) but then BΞ = ∅, a contradiction, since now there is no way to derive ¬BA
suppose A ∈ / Ξ:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Take Γ = {¬BA ⊃ A}. Let Ξ be any classically closed superset of Γ. suppose A ∈ Ξ: but how to ground A? in view of Γ the only way is to have also ¬BA by we have BA ∈ Ξ hence, Ξ is trivial (by classical closure) but then BΞ = ∅, a contradiction, since now there is no way to derive ¬BA
suppose A ∈ / Ξ: then ¬BA
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Extension do not always exist Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Take Γ = {¬BA ⊃ A}. Let Ξ be any classically closed superset of Γ. suppose A ∈ Ξ: but how to ground A? in view of Γ the only way is to have also ¬BA by we have BA ∈ Ξ hence, Ξ is trivial (by classical closure) but then BΞ = ∅, a contradiction, since now there is no way to derive ¬BA
suppose A ∈ / Ξ: then ¬BA but then also A by modus ponens, contradiction Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
99 / 104
Are autoepistemic extensions of some Γ always consistent? Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
100 / 104
Are autoepistemic extensions of some Γ always consistent? Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
If Γ is inconsistent: nope!
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
100 / 104
Are autoepistemic extensions of some Γ always consistent? Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
If Γ is inconsistent: nope! What if Γ is consistent?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
100 / 104
Are autoepistemic extensions of some Γ always consistent? Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
If Γ is inconsistent: nope! What if Γ is consistent? Take Γ = {¬Bp} if p is in we get Bp together with Γ we can derive anything
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
100 / 104
Another Example Take Γ = {¬Bp ⊃ q, ¬Bq ⊃ p}. What are the autoepistemic extensions of Γ? 1
what about one with p and q?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
101 / 104
Another Example Take Γ = {¬Bp ⊃ q, ¬Bq ⊃ p}. What are the autoepistemic extensions of Γ? 1
what about one with p and q? but how to ground p and q?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
101 / 104
Another Example Take Γ = {¬Bp ⊃ q, ¬Bq ⊃ p}. What are the autoepistemic extensions of Γ? 1
what about one with p and q? but how to ground p and q? We would need ¬Bp to ground q
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
101 / 104
Another Example Take Γ = {¬Bp ⊃ q, ¬Bq ⊃ p}. What are the autoepistemic extensions of Γ? 1
what about one with p and q? but how to ground p and q? We would need ¬Bp to ground q and ¬Bq to ground p
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
101 / 104
Another Example Take Γ = {¬Bp ⊃ q, ¬Bq ⊃ p}. What are the autoepistemic extensions of Γ? 1
what about one with p and q? but how to ground p and q? We would need ¬Bp to ground q and ¬Bq to ground p
2
what about one with ¬p and/or ¬q?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
101 / 104
Another Example Take Γ = {¬Bp ⊃ q, ¬Bq ⊃ p}. What are the autoepistemic extensions of Γ? 1
what about one with p and q? but how to ground p and q? We would need ¬Bp to ground q and ¬Bq to ground p
2
what about one with ¬p and/or ¬q? again: how to ground ¬p and/or ¬q?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
101 / 104
Another Example Take Γ = {¬Bp ⊃ q, ¬Bq ⊃ p}. What are the autoepistemic extensions of Γ? 1
what about one with p and q? but how to ground p and q? We would need ¬Bp to ground q and ¬Bq to ground p
2
what about one with ¬p and/or ¬q? again: how to ground ¬p and/or ¬q? only possibility is if Γ were inconsistent: but it isn’t
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
101 / 104
Another Example Take Γ = {¬Bp ⊃ q, ¬Bq ⊃ p}. What are the autoepistemic extensions of Γ? 1
what about one with p and q? but how to ground p and q? We would need ¬Bp to ground q and ¬Bq to ground p
2
what about one with ¬p and/or ¬q? again: how to ground ¬p and/or ¬q? only possibility is if Γ were inconsistent: but it isn’t
3
what about one with p and without q?
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
101 / 104
Another Example Take Γ = {¬Bp ⊃ q, ¬Bq ⊃ p}. What are the autoepistemic extensions of Γ? 1
what about one with p and q? but how to ground p and q? We would need ¬Bp to ground q and ¬Bq to ground p
2
what about one with ¬p and/or ¬q? again: how to ground ¬p and/or ¬q? only possibility is if Γ were inconsistent: but it isn’t
3
what about one with p and without q? then ¬Bq (neg. introspection)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
101 / 104
Another Example Take Γ = {¬Bp ⊃ q, ¬Bq ⊃ p}. What are the autoepistemic extensions of Γ? 1
what about one with p and q? but how to ground p and q? We would need ¬Bp to ground q and ¬Bq to ground p
2
what about one with ¬p and/or ¬q? again: how to ground ¬p and/or ¬q? only possibility is if Γ were inconsistent: but it isn’t
3
what about one with p and without q? then ¬Bq (neg. introspection) thus, p is grounded
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
101 / 104
Another Example Take Γ = {¬Bp ⊃ q, ¬Bq ⊃ p}. What are the autoepistemic extensions of Γ? 1
what about one with p and q? but how to ground p and q? We would need ¬Bp to ground q and ¬Bq to ground p
2
what about one with ¬p and/or ¬q? again: how to ground ¬p and/or ¬q? only possibility is if Γ were inconsistent: but it isn’t
3
what about one with p and without q? then ¬Bq (neg. introspection) thus, p is grounded
4
similarly, one with q and without p
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
101 / 104
Bootstrapping problem (Konolige (1988)) Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
The problem Take Γ = {BA ⊃ A}.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
102 / 104
Bootstrapping problem (Konolige (1988)) Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
The problem Take Γ = {BA ⊃ A}. There is an autoepistemic extension that contains A (and BA and A ∨ B and B(A ∨ B) and ¬B¬A, etc.). Note that A is grounded in view of Γ once we adopt BA.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
102 / 104
Bootstrapping problem (Konolige (1988)) Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
The problem Take Γ = {BA ⊃ A}. There is an autoepistemic extension that contains A (and BA and A ∨ B and B(A ∨ B) and ¬B¬A, etc.). Note that A is grounded in view of Γ once we adopt BA.
Solution
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
102 / 104
Bootstrapping problem (Konolige (1988)) Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
The problem Take Γ = {BA ⊃ A}. There is an autoepistemic extension that contains A (and BA and A ∨ B and B(A ∨ B) and ¬B¬A, etc.). Note that A is grounded in view of Γ once we adopt BA.
Solution Take the minimal autoepistemic extensions!
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
102 / 104
Bootstrapping problem (Konolige (1988)) Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
The problem Take Γ = {BA ⊃ A}. There is an autoepistemic extension that contains A (and BA and A ∨ B and B(A ∨ B) and ¬B¬A, etc.). Note that A is grounded in view of Γ once we adopt BA.
Solution Take the minimal autoepistemic extensions! Note there is another autoepistemic extension Ξ0 of Γ that only contains (introspective) beliefs in classical tautologies and for all non-tautologies C it contains Bi ¬BC (where i ≥ 0). Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
102 / 104
Bootstrapping problem (Konolige (1988)) Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ})
The problem Take Γ = {BA ⊃ A}. There is an autoepistemic extension that contains A (and BA and A ∨ B and B(A ∨ B) and ¬B¬A, etc.). Note that A is grounded in view of Γ once we adopt BA.
Solution Take the minimal autoepistemic extensions! Note there is another autoepistemic extension Ξ0 of Γ that only contains (introspective) beliefs in classical tautologies and for all non-tautologies C it contains Bi ¬BC (where i ≥ 0). If we compare the subsets of Ξ and Ξ0 without occurrences of B, written Ξ0 and Ξ00 , (RUB, thenUGENT) Ξ00 ⊂ Ξ0Tutorial: . Christian Straßer Nonmonotonic Logic (Day 1) September 2, 2015 102 / 104
Bootstrapping problem continued
Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ}) Take Γ = {¬Bp ⊃ q, Bp ⊃ p}
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
103 / 104
Bootstrapping problem continued
Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ}) Take Γ = {¬Bp ⊃ q, Bp ⊃ p} Two (minimal) autoepistemic extensions:
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
103 / 104
Bootstrapping problem continued
Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ}) Take Γ = {¬Bp ⊃ q, Bp ⊃ p} Two (minimal) autoepistemic extensions: 1
including Bq, q and not including p
2
including Bp, p and not including q (and thus including ¬Bq)
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
103 / 104
Bootstrapping problem continued
Recall: Ξ = Cn({Γ ∪ BΞ ∪ ¬BΞ}) Take Γ = {¬Bp ⊃ q, Bp ⊃ p} Two (minimal) autoepistemic extensions: 1
including Bq, q and not including p
2
including Bp, p and not including q (and thus including ¬Bq)
(see Konolige (1988))
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
103 / 104
Topic 1
2 3
4 5
Defeasible Reasoning – Some Basic Concepts Some examples to warm up Some conceptual distinctions Nonmonotonic Logic in Context The Dynamics of Defeasible Reasoning Default Logic (in the tradition of Reiter) Warming up Defaults and Default Theories Inferring Alternatives and more examples Meta-Properties Introducing Priorities Summing up Autoepistemic Logic Bibliography Bibliography
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
104 / 104
Antonelli, G. A.: 1999, ‘A directly cautious theory of defeasible consequence for default logic via the notion of general extension’. Artificial Intelligence 109(1), 71–109. Gabbay, D. M.: 1985, ‘Theoretical foundations for non-monotonic reasoning in expert systems’. In: Logics and models of concurrent systems. New York, NY, USA: Springer-Verlag New York, Inc., pp. 439–457. Gordon, T., H. Prakken, and D. Walton: 2007, ‘The Carneades model of argument and burden of proof’. Artifical Intelligence 171(10), 875–896. Horty, J.: 2007, ‘Defaults with Priorities’. Journal of Philosophical Logic 36, 367–413. Horty, J. F.: 1994, ‘Some direct theories of nonmonotonic inheritance’. In: D. M. Gabbay, C. J. Hogger, and J. A. Robinson (eds.): Handbook of Logic in Artificial Intelligence and Logic Programming, Volume 3: Nonmonotonic Reasoning and Uncertain Reasoning. Oxford University Press, pp. 111–187. Horty, J. F.: 2012, Reasons as defaults. Oxford University Press. Konolige, K.: 1988, ‘On the relation between default and autoepistemic logic’. Artifical Intelligence 35(3), 343–382. Kraus, S., D. Lehman, and M. Magidor: 1990, ‘Nonmonotonic Reasoning, Preferential Models and Cumulative Logics’. Artifical Intelligence 44, 167–207. Łukaszewicz, W.: 1988, ‘Considerations on default logic: an alternative approach’. Computational intelligence 4(1), 1–16. Makinson, D. and P. Gärdenfors: 1991, ‘Relations between the logic of theory change and nonmonotonic logic’. In: The logic of theory change. Springer, pp. 183–205. Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
104 / 104
Moore, R. C.: 1984, ‘Possible-World Semantics for Autoepistemic Logic.’. In: Proceedings of the Workshop on non-monotonic reasoning. pp. 344–354. Moore, R. C.: 1985, ‘Semantical considerations on nonmonotonic logic’. Artifical Intelligence 25(1), 75–94. Pfeifer, N.: 2014, ‘Reasoning about uncertain conditionals’. Studia Logica pp. 1–18. Poole, D.: 1988, ‘A logical framework for default reasoning’. Artificial intelligence 36(1), 27–47. Poole, D.: 1991, ‘The effect of knowledge on belief: conditioning, specificity and the lottery paradox in default reasoning’. Artifical Intelligence 49(1-3), 281–307. Reiter, R.: 1980, ‘A Logic for Default Reasoning’. Artifical Intelligence 1–2(13). Stalnaker, R.: 1993, ‘A note on non-monotonic modal logic’. Artificial Intelligence 64(2), 183–196. Stalnaker, R.: 1994, ‘What is a nonmonotonic consequence relation?’. Fundamenta Informaticae 21(1), 7–21. Stenning, K. and M. Van Lambalgen: 2008, Human reasoning and cognitive science. MIT Press. Toulmin, S. E.: 1958, The Uses of Argument. Cambridge University Press.
Christian Straßer (RUB, UGENT)
Tutorial: Nonmonotonic Logic (Day 1)
September 2, 2015
104 / 104