Sie sind auf Seite 1von 4

Knowledge Representation and Reasoning Collection 2018

Question 1
Consider the following description logic TBox T :

T = {Professor v ∃supervises.DPhilStudent,
DPhilStudent v Student,
∃supervises.(DPhilStudent u Student) v Supervisor}.

(a) Show that T 6|= Professor v ∀supervises.(DPhilStudent u ¬Student). (3 marks)

(b) Using the tableau algorithm for ALC show that T |= Professor v Supervisor (you can use
any optimisation technique you know). (7 marks)

(c) Describe how the EL consequence-based algorithm can be used in order to show that
T 6|= Professor v ∃supervises.(DPhilStudent u ∃supervises.Student). (You do not need to
specify the detailed execution of the algorithm; i.e., you do not need to specify which rules
are actually applied. You are, however, required to specify the precise input that would
be given to the consequence-based algorithm.) (4 marks)

(d) Describe the relationship between the description logic ALC and the following logics:

• the description logic EL;


• the Guarded Fragment of First Order Logic;
• the Two-variable Fragment of First Order Logic.

(3 marks)

(e) Reduce the problem of checking whether T |= Professor v Supervisor to checking unsatis-
fiability of a set of clauses S. What is the Herbrand Universe of S? (4 marks)

(f) Construct a union of CQs rewriting of the Boolean CQ

q = A(x) ∧ B(x) ∧ R(x, y)

and the flat DL-Lite TBox


B v A,
∃P.> v B.
(4 marks)

–1– TURN OVER


Question 2
(a) Give the formal definition of the least Herbrand model. (3 marks)

(b) Let disjunctive Datalog be the language that extends Datalog with a possibility of dis-
junction in the head (i.e., on the right hand side) of a rule. Does a knowledge base in
disjunctive Datalog always have a least Herbrand model? Justify your answer formally,
that is, provide a proof (e.g., counterexample). (3 marks)

(c) Using N and natural numbers as constants, unary predicates Y ear, Leap and Common,
as well as binary predicate DivisibleBy, formalise as a Datalog¬ knowledge base K the
following:

- a year is leap if it is either divisible by 400, or it is divisible by 4 and not divisible by


100;
- if a year is not leap, then it is common;
- N is a year.

(5 marks)

(d) Consider the Herbrand interpretations I1 and I2 such that

Y earI1 = {N }, CommonI1 = {N }, LeapI1 = ∅, DivisibleBy I1 = ∅;


Y earI2 = {N }, CommonI2 = {N }, LeapI2 = ∅, DivisibleBy I2 = {(N, 4), (N, 100)}.

Which of these interpretations are stable models of K? Justify your answer formally.
Are your answers enough to conclude that K |≈ Common(N )? If not, what else should
we check to prove that K |≈ Common(N )? (4 marks)

(e) Give an example of a Datalog ¬ knowledge base without a stable model. (3 marks)

(f) - Give an example of a problem that is in NP, but known to be not NP-complete.
- Give an example of a problem that is NP-hard, but known to be not NP-complete.
Justify your answers. (3 marks)

(g) Consider the following fragment of FO, which generalises both ALC and Datalog.
An ALC+Datalog knowledge base consists of

- ALC TBox T ,
- a set of positive (i.e., without ⊥ in the head) Datalog rules R,
- a set of facts A,

such that the set of predicate symbols (i.e., concepts and roles) of T and set of predicate
symbols of R are disjoint. Is the KB satisfiability problem in ALC+Datalog decidable?
Justify your answer, that is, either describe an algorithm or explain why it does not exist.
(4 marks)

–2–
Question 3
(a) Consider unary predicates Employee, Department, Manager, and CEO as well as binary
predicates: WorksFor, HeadOf, and Manages. Using this vocabulary, write down a first-
order sentence for representing each of the following English statements:

• Every employee works for some department.


• A CEO is a manager that manages all employees who are not CEOs themselves.
• If an employee works for a department and is not head of that department, then that
employee is not a manager.
• Only managers can be head of a department.
• Every manager is an employee.
• Every employee either manages some other employee, or is managed by some other
employee.

(3 marks)

(b) Let K be the set of first order sentences you have written in Part (a), and let

K0 = K ∪ {Manager(peter), Manages(peter, john),


WorksFor(john, f inance), Department(f inance)}.

Determine whether the following entailment holds (justify your answer):

K0 |= ∃x.(Department(x) ∧ WorksFor(peter, x)).

(2 marks)

(c) Define the Bernays-Schönfinkel fragment of first order logic. Determine which of the
first order sentences you have just written in Part (a) belong to the Bernays-Schönfinkel
fragment. (4 marks)

(d) Give the definition of finite model property for a fragment of first-order logic? Show that
the Bernays-Schönfinkel fragment possesses this property. (4 marks)

(e) Explain which of the following sentences are in NNF and which are not:

α1 = ∀x.∃y.(R(x, y) → A(y)),
α2 = ¬(∀x.∃y.R(x, y)).

For sentences that are not in NNF construct equivalent ones in NNF. (3 marks)

(f) Skolemise the formulas in NNF equivalent to α1 and α2 that you have written in Part (d)
(you should Skolemise even if you found α1 or α2 to be already in NNF).
Let us call the corresponding Skolemised formulas α100 and α200 , respectively.

• Does α100 entail α1 ? Prove that your answer is correct.


• Does α2 entail α200 ? Prove that your answer is correct.

–3– TURN OVER


(4 marks)

(g) Consider the first-order logic sentence α = ∀x1 . . . ∀xn .∃y.φ(x1 , . . . , xn , y) where n ≥ 1,
φ(x1 , . . . , xn , y) is a quantifier-free first-order logic formula in NNF, and φ contains only
variables x1 , . . . , xn and y. Let α0 be the result of Skolemising α.

• Does α0 entail α? If you think it does, provide a proof; otherwise, provide a counter-
example.

(5 marks)

–4– LAST PAGE

Das könnte Ihnen auch gefallen