• Keine Ergebnisse gefunden

Equivalence of Level-0: PBDL and OWP

2.3 Deontic Action Logics

2.3.4 Equivalence of Level-0: PBDL and OWP

Now I provide a correspondence from PBDL and OWP, and this result shows that PBDL does partially overlap OWP. In other words, a fragment of the language for PBDL is a translation of a fragment for OWP, and vice versa. The corresponding construction is similar to the method introduced in [77].

I start with a syntactic translation from a fragment of PBDL to a fragment of OWP in two steps. First I define a fragment of PBDL in which no dynamic sentences and weak permission occur. Let L be the fragment of language for PBDL, which in-volves formulas containing atomic propositions, action generators, and the formulas constructed with Boolean connectives, dynamic operator, and strong permission. To

2.3. Deontic Action Logics 35

Table 2.4: The sound and complete axiom system for Castro and Maibaum’s Propo-sitional Boolean Dynamic Logic. All propoPropo-sitional tautologies, Modus Ponens, and the axioms for Boolean Algebras for action terms are also taken into account. Above γAct0, andϕ[α/β]indicates that all occurrences ofαinϕare replaced byβ.

construct the translation fragment of PBDL, I allow for dynamic operators and per-missions to scope over formulas ofL. I simply call this fragmentL×. LetL be the language for OWP. The translation proceeds in two steps. Define σLLbe a translation as follows:

The resulting δPWα) embodies WRO. So α is a type of strategy required to be played, iff all rationally permissible strategies areα-type strategies.

Let M = ⟨W, E, R, I, RP,∣∣⋅∣∣⟩ be a PBDL model. I construct a modelM =

H, nP, nO,∣∣⋅∣∣⟩as follows:

H =WE.

nP(h)={ {XE ∶ ∀eX. RPhe} ifhW

∅ ifhE

nO =∅

• ∣∣σ(a)∣∣ =I(a)

• ∣∣σ(p)∣∣ =∣∣p∣∣

I denote M as the unique model generated from a PBDL model M. It should be noticed thatM is an OWP model.

2.3.11.PROPOSITION. Given aPBDLmodelM =⟨W, E, R, I, RP,∣∣⋅∣∣⟩,Mis an OWPmodel.

Proof:

The condition fornP is easy to verify by the construction. AsnO is empty, the three conditions required for OWP models are automatically satisfied. □

I now can show one of the main results in this section.

2.3.12.THEOREM. Given a PBDL model M = ⟨W, E, R, I, RP,∣∣ ⋅∣∣⟩. Then, for eachα, ϕL, andψL×,

eI(α) if and only if e∈∣∣σ(α)∣∣ M, wϕ if and only if M, wσ(ϕ) M, wψ only if M, wδ(ψ) Proof:

It is not difficult to provee ∈ ∣∣σ(α)∣∣ iff eI(α). Now I am going to prove that M, wϕiffM, wσ(ϕ)for eachϕL. I do so by induction on the complexity ofϕL.

1. The cases of atomic propositions, negation, implication are easy to verify by the construction.

2. The case ofαβ:

M, wαβiffI(α)=I(β)

iffe∈∣∣σ(α)∣∣e∈∣∣σ(β)∣∣ iffM, w ⊨□(σ(α)↔σ(β))

2.3. Deontic Action Logics 37 3. The case ofP α: According to the aforementioned construction, then

M, wP αiff∀eI(α). RPwe iff∣∣σ(α)∣∣nP(w)

Next I will verify thatM, wψM, wδ(ψ)for eachψL×. I show this by induction on the complexity ofϕL×.

1. The case of⟨αϕ:

M, w ⊨⟨αϕiff∃(w, u, e)∈RandeI(α)s.t.M, uϕ

only if∃w, e, uW s.t.e∈∣∣σ(α)∣∣andu∈∣∣δ(ϕ)∣∣ iffM, w ⊨◇(σ(α)∧ ◇δ(ϕ))

2. The case of¬PWα:

M, w ⊨¬PWαiff¬∃eI(α)s.t.RPwe

iff∀eE. (RPweeI(α))

only if∀βAct.eE. ((eI(β)⇒ RPwe)

⇒ (eI(β)⇒ eI(α)))

iff∀βAct.[∣∣σ(β)∣∣nP(w)⇒ ∣∣σ(β)∣∣ ⊆∣∣σ(α)∣∣] iffM, wP σ(β)→□(σ(β)→σ(α)), for allβAct iffM, w ⊨ ⋀

βAct

[P σ(β)→□(σ(β)→σ(α))]

Now I can start with the second contribution in this section, which is a correspon-dence result from a fragment of OWP to a fragment of PBDL. Recall that OWP and MDL share the common language. So given the language L for PBDL, similarly, I simply call the fragment of OWP where no modal operator (□, O or P) occurs L0, and the fragment, the so-calledL1of OWP, is defined within the classical conjunction, negation, and modal operators based onL0. Again, sinceAct0is finite, my translation only considers finitely many atomic propositions. I show thatL1can be translated back to PBDL in a non-trivial way. Similarly, the translation runs in two steps.

LetL0be a fragment of OWP in which neither□,O, norP occur. Define a function µL0L+as an action-translation as follows:

µ(p)∈Act0for eachpP rop0 µϕ)∶=µ(ϕ)

µ(ϕψ)∶=µ(ϕ)∩µ(ψ)

I can then define a translation fromL1 toL+. The translationL1L+is defined for complex formulas, which relies on the action-translationµdefined above.

(p)∈P rop0 for eachpP rop0 ϕ)∶=¬(ϕ)

(ϕψ)∶=(ϕ)∧(ψ) (◇ϕ)∶=⟨µ(ϕ)⟩⊤

(P ϕ)∶=P µ(ϕ)

()∶=P µ(ϕ)∧ ¬PWµϕ)

Notice that obligation in OWP can be translated by a combination of strong permission and weak permission in PBDL via the translation.

The transformation from the OWP models to PBDL models need certain restric-tions. Not every OWP model can be transformed into a PBDL model. To do so, the transformation needs to satisfy two conditions. The first is the valuation of atomic propositions in the transformed models should satisfy the required conditions (I.1)-(I.3). Second, the domain of the transformed model should be restricted to the valua-tion of atomic proposivalua-tions in OWP.

To satisfy the first restriction, I define the following “functional OWP models.” An OWP model MOWP = ⟨H, nP, nO,∣∣⋅∣∣⟩is functional, if and only if it satisfies the following three conditions:

• For eachpP rop0,∣∣∣p∣∣−⋃{∣∣q∣∣∶q∈(P rop0−{p})}∣ ≤1

• For eachwW, ifw∈∣∣p∣∣∩∣∣q∣∣such thatpqP rop0, then

⋂{∣∣p∣∣∶pP rop0andw∈∣∣p∣∣} ={w}

W =⋃pP rop0∣∣p∣∣

The three conditions correspond to the conditions (I.1)-(I.3) in PBDL models.

I now can construct a model transformed from a functional OWP model. Let MOWP =⟨H, nP, nO,∣∣⋅∣∣⟩be a functional OWP model, I then can construct a model M=⟨W, E, R, I, RP,∣∣⋅∣∣⟩as follows:

W =E =H

R={(x, y, y) ∣ (x, y)∈Alt}

RP ={(x, y) ∣∃XnP(x)s.t. yX}

I(µ(p))=∣∣p∣∣

• ∣∣(p)∣∣=∣∣p∣∣for eachpP rop0 in OWP-language

2.3. Deontic Action Logics 39 I denoteM as the unique model generated from a functional OWP modelMOWP. 2.3.13.PROPOSITION. Given a functional OWPmodelMOWP = ⟨H, nP, nO,∣∣⋅∣∣⟩, then the modelMdefined above is aPBDLmodel.

Proof:

I only need to check whether R is functional. BecauseMOWP is functional, it is thus

easy to check thatRinMis also functional. □

With this result in hand, I now can turn to the final step of my transformation.

2.3.14.THEOREM. Given a functionalOWP modelM = ⟨H, nP, nO,∣∣⋅∣∣⟩, I con-struct the modelM =⟨W, E, R, I, RP,∣∣⋅∣∣⟩generated fromM. Then, given any OWP sentenceϕandh∈⋃pP ropo0∣∣p∣∣,

1. IfϕL0, thenM, hϕiffhI(µ(ϕ)) 2. IfϕL1, thenM, hϕiffM, h(ϕ) Proof:

1. SupposeϕL0. By induction on the complexity ofϕ.

(a) For the case ofpP rop0, then

M, hpiffh∈∣∣p∣∣iffhI(µ(p))

(b) For the case ofϕ=ψϕL0, then alsoψ, ϕL0. Now I have M, hψϕiffM, hψandM, hϕ

iffhI(µ(ψ))andhI(µ(ϕ)) iffhI(µ(ψ))∩I(µ(ϕ)) iffhI(µ(ψ)∩µ(ϕ)) iffhI(µ(ψϕ))

(c) For the case of¬ϕL0, then alsoϕL0. We can see that M, h⊨¬ϕiffM, hϕ

iffh∈/ I(µ(ϕ)) iffhI(µ(ϕ)) iffhI(µϕ)) 2. SupposeϕL1. By induction on the complexity ofϕ.

(a) For the case ofpP rop0, then

M, hpiffh∈∣∣p∣∣iffh ∈∣∣(p)∣∣iffM, h(p)

(b) For the case ofψϕL1, then alsoψ, ϕL1. I can then infer M, hψϕiffM, hψ andM, hϕ

iffM, h(ψ)andM, h(ϕ) iffM, h(ψ)∧(ϕ)

iffM, h(ψϕ)

(c) For the case of¬ψL1, then alsoψL1. So I infer M, h⊨¬ψiffM, hψ

iffM, h(ψ) iffM, hψ)

(d) For the case of◇ψL1, then alsoψL0by the definition. So M, h⊨◇ψ iff∃hHs.t. Alt(h, h)andM, hψ

iff∃hI(µ(ψ))s.t.Rh(h)=handM, h ⊨ ⊤ iffM, h⊨⟨µ(ψ)⟩⊤

iffM, h(◇ψ)

(e) Similar to the case ofP ψL1, thenψL0. So M, hP ψiff∣∣ψ∣∣∈nP(h)

iff∀h ∈∣∣ψ∣∣thatRPhh by the Construction ofRP iff∀hI(µ(ψ))thatRPhh

iffM, hP µ(ψ) iffM, h(P ψ)

2.4. Conclusion 41