15.08.2013 Views

General Computer Science 320201 GenCS I & II Lecture ... - Kwarc

General Computer Science 320201 GenCS I & II Lecture ... - Kwarc

General Computer Science 320201 GenCS I & II Lecture ... - Kwarc

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

A ∨ BT AT <br />

<br />

BT A ⇒ BT AF <br />

<br />

BT A ∨ B F<br />

A F<br />

B F<br />

A ⇒ B F<br />

A T<br />

B F<br />

A ⇔ BT AT BT <br />

<br />

<br />

AF<br />

BF A T<br />

A ⇒ B T<br />

B T<br />

A ⇔ BF AT BF <br />

<br />

<br />

AF<br />

BT c○: Michael Kohlhase 205<br />

A T<br />

A ⇒ B T<br />

¬A ∨ B T<br />

¬(¬¬A ∧ ¬B) T<br />

¬¬A ∧ ¬B F<br />

¬¬A F<br />

¬A T<br />

A F<br />

⊥<br />

With these derived rules, theorem proving becomes quite efficient. With these rules, the tableau<br />

(??) would have the following simpler form:<br />

Tableaux with derived Rules (example)<br />

Example 343<br />

love(mary, bill) ∧ love(john, mary) ⇒ love(john, mary) F<br />

love(mary, bill) ∧ love(john, mary) T<br />

love(john, mary) F<br />

love(mary, bill) T<br />

love(john, mary) T<br />

⊥<br />

c○: Michael Kohlhase 206<br />

Another thing that was awkward in (??) was that we used a proof for an implication to prove<br />

logical consequence. Such tests are necessary for instance, if we want to check consistency or<br />

informativity of new sentences 11 . Consider for instance a discourse ∆ = D 1 , . . . , D n , where n is EdNote:11<br />

large. To test whether a hypothesis H is a consequence of ∆ (∆ |= H) we need to show that<br />

C := (D 1 ∧ . . .) ∧ D n ⇒ H is valid, which is quite tedious, since C is a rather large formula, e.g.<br />

if ∆ is a 300 page novel. Moreover, if we want to test entailment of the form (∆ |= H) often, –<br />

for instance to test the informativity and consistency of every new sentence H, then successive<br />

∆s will overlap quite significantly, and we will be doing the same inferences all over again; the<br />

entailment check is not incremental.<br />

Fortunately, it is very simple to get an incremental procedure for entailment checking in the<br />

model-generation-based setting: To test whether ∆ |= H, where we have interpreted ∆ in a model<br />

generation tableau T , just check whether the tableau closes, if we add ¬H to the open branches.<br />

Indeed, if the tableau closes, then ∆ ∧ ¬H is unsatisfiable, so ¬((∆ ∧ ¬H)) is valid 12 , but this is EdNote:12<br />

equivalent to ∆ ⇒ H, which is what we wanted to show.<br />

¬B F<br />

B T<br />

Example 344 Consider for instance the following entailment in natural language.<br />

Mary loves Bill. John loves Mary |= John loves Mary<br />

13 We obtain the tableau EdNote:13<br />

love(mary, bill) T<br />

love(john, mary) T<br />

¬(love(john, mary)) T<br />

love(john, mary) F<br />

⊥<br />

11 EdNote: add reference to presupposition stuff<br />

12 EdNote: Fix precedence of negation<br />

13 EdNote: need to mark up the embedding of NL strings into Math<br />

110

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!